WO2002069200A1 - Checkpointing filesystem - Google Patents

Checkpointing filesystem Download PDF

Info

Publication number
WO2002069200A1
WO2002069200A1 PCT/US2002/005613 US0205613W WO02069200A1 WO 2002069200 A1 WO2002069200 A1 WO 2002069200A1 US 0205613 W US0205613 W US 0205613W WO 02069200 A1 WO02069200 A1 WO 02069200A1
Authority
WO
WIPO (PCT)
Prior art keywords
filesystem
user data
checkpointing
wfat
dfat
Prior art date
Application number
PCT/US2002/005613
Other languages
French (fr)
Inventor
Alan G. Gara
Mark E. Giampapa
Burkhard D. Steinmacher-Burow
Original Assignee
Gara Alan G
Giampapa Mark E
Steinmacher-Burow Burkhard D
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Gara Alan G, Giampapa Mark E, Steinmacher-Burow Burkhard D filed Critical Gara Alan G
Priority to PCT/US2002/005613 priority Critical patent/WO2002069200A1/en
Priority to US10/258,515 priority patent/US6895416B2/en
Publication of WO2002069200A1 publication Critical patent/WO2002069200A1/en

Links

Classifications

    • HELECTRICITY
    • H05ELECTRIC TECHNIQUES NOT OTHERWISE PROVIDED FOR
    • H05KPRINTED CIRCUITS; CASINGS OR CONSTRUCTIONAL DETAILS OF ELECTRIC APPARATUS; MANUFACTURE OF ASSEMBLAGES OF ELECTRICAL COMPONENTS
    • H05K7/00Constructional details common to different types of electric apparatus
    • H05K7/20Modifications to facilitate cooling, ventilating, or heating
    • H05K7/20709Modifications to facilitate cooling, ventilating, or heating for server racks or cabinets; for data centers, e.g. 19-inch computer racks
    • H05K7/20836Thermal management, e.g. server temperature control
    • FMECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F04POSITIVE - DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS FOR LIQUIDS OR ELASTIC FLUIDS
    • F04DNON-POSITIVE-DISPLACEMENT PUMPS
    • F04D25/00Pumping installations or systems
    • F04D25/16Combinations of two or more pumps ; Producing two or more separate gas flows
    • F04D25/166Combinations of two or more pumps ; Producing two or more separate gas flows using fans
    • FMECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F04POSITIVE - DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS FOR LIQUIDS OR ELASTIC FLUIDS
    • F04DNON-POSITIVE-DISPLACEMENT PUMPS
    • F04D27/00Control, e.g. regulation, of pumps, pumping installations or pumping systems specially adapted for elastic fluids
    • F04D27/004Control, e.g. regulation, of pumps, pumping installations or pumping systems specially adapted for elastic fluids by varying driving speed
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00Error detection; Error correction; Monitoring
    • G06F11/07Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/14Error detection or correction of the data by redundancy in operation
    • G06F11/1402Saving, restoring, recovering or retrying
    • G06F11/1415Saving, restoring, recovering or retrying at system level
    • G06F11/1438Restarting or rejuvenating
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F15/00Digital computers in general; Data processing equipment in general
    • G06F15/16Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
    • G06F15/163Interprocessor communication
    • G06F15/173Interprocessor communication using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
    • G06F15/17356Indirect interconnection networks
    • G06F15/17368Indirect interconnection networks non hierarchical topologies
    • G06F15/17381Two dimensional, e.g. mesh, torus
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/14Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/141Discrete Fourier transforms
    • G06F17/142Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46Multiprogramming arrangements
    • G06F9/52Program synchronisation; Mutual exclusion, e.g. by means of semaphores
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46Multiprogramming arrangements
    • G06F9/52Program synchronisation; Mutual exclusion, e.g. by means of semaphores
    • G06F9/526Mutual exclusion algorithms
    • GPHYSICS
    • G09EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
    • G09GARRANGEMENTS OR CIRCUITS FOR CONTROL OF INDICATING DEVICES USING STATIC MEANS TO PRESENT VARIABLE INFORMATION
    • G09G5/00Control arrangements or circuits for visual indicators common to cathode-ray tube indicators and other visual indicators
    • G09G5/003Details of a display terminal, the details relating to the control arrangement of the display terminal and to the interfaces thereto
    • G09G5/006Details of the interface to the display terminal
    • G09G5/008Clock recovery
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L7/00Arrangements for synchronising receiver with transmitter
    • H04L7/02Speed or phase control by the received code signals, the signals containing no special synchronisation information
    • H04L7/033Speed or phase control by the received code signals, the signals containing no special synchronisation information using the transitions of the received signal to control the phase of the synchronising-signal-generating means, e.g. using a phase-locked loop
    • H04L7/0337Selecting between two or more discretely delayed clocks or selecting between two or more discretely delayed received code signals
    • H04L7/0338Selecting between two or more discretely delayed clocks or selecting between two or more discretely delayed received code signals the correction of the phase error being performed by a feed forward loop
    • FMECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F24HEATING; RANGES; VENTILATING
    • F24FAIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
    • F24F11/00Control or safety arrangements
    • F24F11/70Control systems characterised by their outputs; Constructional details thereof
    • F24F11/72Control systems characterised by their outputs; Constructional details thereof for controlling the supply of treated air, e.g. its pressure
    • F24F11/74Control systems characterised by their outputs; Constructional details thereof for controlling the supply of treated air, e.g. its pressure for controlling air flow rate or air velocity
    • F24F11/77Control systems characterised by their outputs; Constructional details thereof for controlling the supply of treated air, e.g. its pressure for controlling air flow rate or air velocity by controlling the speed of ventilators
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02BCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
    • Y02B30/00Energy efficient heating, ventilation or air conditioning [HVAC]
    • Y02B30/70Efficient control or regulation technologies, e.g. for control of refrigerant flow, motor or heating
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10TECHNICAL SUBJECTS COVERED BY FORMER USPC
    • Y10STECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10S707/00Data processing: database and file management or data structures
    • Y10S707/99951File or database maintenance
    • Y10S707/99952Coherency, e.g. same view to multiple users
    • Y10S707/99953Recoverability
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10TECHNICAL SUBJECTS COVERED BY FORMER USPC
    • Y10STECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10S707/00Data processing: database and file management or data structures
    • Y10S707/99951File or database maintenance
    • Y10S707/99952Coherency, e.g. same view to multiple users
    • Y10S707/99955Archiving or backup

Definitions

  • the present invention generally relates to a field of f ⁇ lesystems in computers. More particularly, the present invention is directed to efficiently checkpointing a filesystem on a distributed-memory parallel supercomputer, thereby facilitating faster execution of applications.
  • large computing systems such as a distributed-memory parallel supercomputer
  • checkpointing the application mitigates the rerunning of the application since the application need only be rolled back to the previous checkpoint rather than be rerun from the start.
  • checkpointing of the filesystem is a critical aspect of checkpointing in the large computing systems, such as the distributed-memory parallel supercomputer.
  • a main motivation for a distributed-memory parallel supercomputer is a fast execution of the application.
  • a checkpointing filesystem which is not significantly slower than a filesystem without checkpointing.
  • a checkpointing filesystem in which the act of checkpointing is fast, since during checkpointing of the filesystem the application is not executing.
  • a checkpointing filesystem which appears to the application as a normal filesystem and does not complicate the implementation of the application.
  • a method for checkpointing a filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: informing the node and the interface to complete their access to the filesystem in order to take a checkpoint of the filesystem; directing the interface to take the checkpoint of the associated disk, the taking of the checkpoint comprising a step of clearing: (i) a recently written file allocation table (WFAT) for maintaining information regarding the user data written to the disk since a previously taken checkpoint; (ii) a recently deleted file allocation table (DFAT) for maintaining information regarding the user data deleted from the disk since the previously taken checkpoint; and informing the node and the interface that they may resume their access to the filesystem, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
  • WFAT recently written file allocation table
  • DFAT recently deleted file allocation
  • a method for rolling back a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem to a previously taken checkpoint of the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data
  • the method comprising the steps of: informing the node and the interface to complete their access to the filesystem in order rollback to the previously taken checkpoint of the filesystem; directing the interface to rollback to the previously taken checkpoint of the associated disk, the rolling back to the previously taken checkpoint comprising: (i) removing user data from a file allocation table (FAT) utilizing a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since the previously taken checkpoint; (ii) adding user data to the FAT from a recently deleted file allocation table (DFAT) that maintains information regarding user data deleted from the disk since the previously taken checkpoint; and informing the node and the interface that they
  • a method for writing user data to a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses the user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data
  • the method comprising the steps of: selecting sectors that are marked as free in a file allocation table (FAT) and not marked as deleted in a recently deleted file allocation table (DFAT) that maintains information regarding user data deleted from the disk since a previously taken checkpoint; writing the user data to the selected sectors on the disk; updating the FAT to reflect the written user data; and updating a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
  • FAT file allocation table
  • DFAT recently deleted file allocation table
  • a method for deleting user data from a checkpointing filesystem of a distributed- memory parallel supercomputer comprising a node that accesses the user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data
  • the method comprising the steps of: deleting sectors associated with the user data to be deleted from the disk from a file allocation table (FAT); determining whether the sectors to be deleted are marked in a recently written file allocation table (WFAT) that maintains information regarding user data written to the disk since a previously taken checkpoint; and updating a recently deleted file allocation table (DFAT) if the sectors are not marked in WFAT at step (b), the DFAT maintaining information regarding the user data deleted from the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
  • WFAT recently written file allocation table
  • a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, comprising: a host controller comprising means for informing the node and the interface to complete their access to the filesystem in order to take a checkpoint of the filesystem, directing the interface to take a checkpoint of the associated disk, and informing the node and the interface that they may resume their access to the filesystem after the checkpoint it taken; and the interface comprising a means for clearing: a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since a previously taken checkpoint; and a recently deleted file allocation table (DFAT) for maintaining information regarding the user data deleted from the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
  • WFAT recently written file allocation table
  • DFAT recently deleted file allocation table
  • the checkpointing filesystem of the distributed-memory parallel supercomputer further comprises rolling back to the previously taken checkpoint, wherein: the host controller further comprises means for informing the node and the interface to complete their access to the filesystem in order rollback to the previously taken checkpoint of the filesystem, for directing the interface to rollback to the previously taken checkpoint of the associated disk, and for informing the node and the interface that they may resume their access to the filesystem; and the interface further comprises means for rolling back to the previously taken checkpoint, the means comprising removing user data from a file allocation table (FAT) utilizing the WFAT and adding user data to the FAT from DFAT.
  • FAT file allocation table
  • the checkpointing filesystem of the distributed-memory parallel supercomputer further comprising writing user data to the checkpointing filesystem, wherein: the interface further comprises means for selecting sectors that are marked as free in the FAT and not marked for deletion in DFAT, for writing the user data to the selected sectors on the disk; for updating the FAT to reflect the written user data, and for updating WFAT.
  • the checkpointing filesystem of a distributed-memory parallel supercomputer further comprising deleting user data from the checkpointing filesystem, wherein: the interface further comprises means for deleting sectors associated with the user data to be deleted from the disk from the FAT, for determining whether the sectors to be deleted are marked in the WFAT, and for updating the DFAT if the sectors are not marked in WFAT.
  • Figure 1 illustrates an exemplary high-level system diagram for performing checkpointing of the filesystem according to the present invention
  • Figure 2 illustrates is an exemplary more-detailed system diagram of Figure 1 for performing checkpointing of the filesystem according to the present invention
  • Figure 3 illustrates a conventional filesystem without checkpointing
  • Figure 4 illustrates an exemplary checkpointing filesystem according to the present invention
  • Figure 5 illustrates an exemplary method flowchart that depicts taking a checkpoint in the checkpointing filesystem of Figure 4 according to the present invention
  • Figure 6 illustrates an exemplary method flowchart that depicts rolling back to a previously taken checkpoint in the checkpointing filesystem of Figure 4 according to the present invention
  • Figure 7 illustrates an exemplary method flowchart that depicts writing of user data to the checkpointing filesystem 400 of Figure 4 according to the present invention
  • Figure 8 illustrates an exemplary method flowchart that depicts deleting of user data from the checkpointing filesystem 400 according to the present invention.
  • Figure 9 illustrates an exemplary flowchart that depicts reading of user data in the checkpointing filesystem 400 according to the present invention.
  • the present invention is directed a checkpointing filesystem on the distributed- memory parallel supercomputer that facilitates faster execution of an application executing on the distributed-memory parallel supercomputer.
  • FIG 1 is an exemplary high-level system diagram 100 for performing checkpointing of the filesystem according to the present invention.
  • the high performance computer engine (i.e., "HPCE") 102 represents the distributed- memory parallel supercomputer.
  • the HPCE is a distributed-memory parallel supercomputer that comprises a plurality of nodes.
  • the HPCE 102 communicates via the network switch 104 with the fileserver/filesystem 106 and host control server 108.
  • the HPCE communicates with the foregoing system components over communication connections 110, which may represent standard networking connections, such as Ethernet and the like.
  • the host control server 108 controls the HPCE 102 and the fileserver/filesystem 106, and determines when to perform a checkpoint of the filesystem and when to restore the filesystem from a previous checkpoint as will be described below in greater detail.
  • Figure 2 is an exemplary more-detailed system diagram 200 of Figure 1 for performing checkpointing of the filesystem according to the present invention.
  • the fileserver/filesystem 106 comprises a plurality of personal computers (i.e., "PCs") 202.
  • the distributed-memory parallel supercomputer comprises approximately 1024 PCs.
  • each PC 202 serves a Redundant Array of Independent Disks (i.e., "RAID") cage, i.e., a plurality of disk drives.
  • RAID Redundant Array of Independent Disks
  • each PC 202 serves as an interface for the HPCE 102 to an associated disk 204 via the network switch 104.
  • the HPCE 102 communicates with the fileserver/filesystem 106 via messages that represent basic file operations, such as read, write and delete.
  • a node of the HPCE 102 may access any file stored on any disk 204 via associated PC interface 202.
  • the host control 108 informs each of the nodes of the HPCE 102 and the PCs 202 to complete any disk activity (i.e., access) and not to start any new disk activity.
  • the host control 108 directs each PC 202 to make a checkpoint of its associated disk 204.
  • This checkpointing activity is completely local to each PC 202 and its associated disk 204.
  • the checkpointing activity is the same with regard to all PCs, i.e., as if a single PC 202 performs a checkpoint of its associated disk 204.
  • the host control 108 is a complete computer including its own internal filesystem, which maintains the software for driving the host control 108, including the taking of a checkpoint and rolling back to previous checkpoint as described below in greater detail.
  • each PC 202 is a complete computer including its own internal filesystem, in addition to the associated disk 204.
  • the internal filesystem of the PC 202 maintains the software for driving the PC 202, including the taking of a.
  • FIG. 3 depicts a conventional filesystem 300 without checkpointing.
  • the filesystem 300 occupies an entire disk or a portion of the disk.
  • the filesystem comprises two parts.
  • the first part is user data 304 and the free space 306 in the filesystem.
  • the second part is a File Allocation Table (i.e., "FAT"), also known as metadata, which includes a description of the user data 304 and the free space 306.
  • FAT 302 includes filenames, file data start sectors, file sizes, as well as free sectors.
  • An example of such a FAT 302 is the industry standard FAT32.
  • FIG 4 is an exemplary checkpointing filesystem 400 according to the present invention.
  • the filesystem comprises five parts.
  • the first part is the user data 304 and the free space 306 in the checkpointing filesystem 400, which are analogous to the filesystem without checkpointing 300 in Figure 3.
  • the second part is the File Allocation Table (i.e., "FAT") 302, which is analogous to that of the filesystem without checkpointing 300 in Figure 3.
  • the third part of the checkpointing filesystem 400 is a recently Written FAT (i.e., "WFAT”) 402, which is structured like the FAT 302, but only maintains information on the user data 304 written since a previous checkpoint.
  • FAT File Allocation Table
  • the fourth part of the checkpointing filesystem 400 is a recently Deleted FAT (i.e., "DFAT") 404, which is structured like the FAT 302, but only maintains information on the user data 304 deleted since the previous checkpoint.
  • the fifth and final part the checkpointing filesystem 400 is the recently Written or Deleted list (i.e., "WDLIST") of user data, which includes the same information as the WFAT 402 and the DFAT 404.
  • the WFAT 402 and DFAT 404 are optimal for fast random access, since they provide indexed by sector access to information about recently written or deleted user data 304.
  • the WFAT 402 includes an entry for every sector in the checkpointing filesystem 400.
  • FIG. 5 is an exemplary method flowchart 500 that depicts taking a checkpoint in the checkpointing filesystem 400 according to the present invention. The taking of the checkpoint starts at step 502.
  • the WFAT 402 is cleared, since at the checkpoint the user data 304 written since a previous checkpoint becomes part of the checkpoint. Since the WFAT 402, just like the FAT 302, is a large data structure, clearing the WFAT 402 by iterating through the entire data structure is inefficient. Because the WDLIST 406 is a much smaller data structure than the WFAT 402, it is faster to clear the WFAT 402 using entries in the WDLIST 406.
  • the DFAT 404 is cleared since at the checkpoint the user data 304 deleted since the previous checkpoint must be deleted from the checkpoint. For efficiency, the DFAT 404 is cleared using WDLIST 406.
  • the WDLIST 406 is a list of all sectors in the WFAT 402 and DFAT 404 marking user data 304 written or deleted since the previous checkpoint.
  • the WFAT 404 and DFAT 404 are cleared by unmarking each sector listed in the WDLIST 406.
  • the WDLIST 406 is also cleared by deleting its list of marked sectors.
  • the filesystem is in a clean or consistent state.
  • the exemplary taking of the checkpoint 500 is fast as required by the distributed-memory parallel supercomputer illustrated in Figures 1 and 2, and minimally impacts the application.
  • FIG. 6 is an exemplary method flowchart 600 that depicts rolling back to a previously taken checkpoint according to the present invention.
  • the rolling back to the previously taken checkpoint starts at step 602.
  • the FAT 302 is returned to the previously taken checkpoint using WDLIST 406. More particularly, at the rollback to the previously taken checkpoint, the user data 304 written since the previous checkpoint is removed from the FAT 302, and the user data 304 deleted since the previous checkpoint is returned to the FAT 302. Since the WDLIST 406 includes all the information on the user data 304 written and deleted since the previously taken checkpoint, the WDLIST is used to return the FAT to the previously taken checkpoint.
  • the WFAT 402 is cleared utilizing WDLIST 406 for efficiency.
  • the WDLIST 406 is a list of all sectors in the WFAT 402 and DFAT 404 marking user data 304 written or deleted since the previous checkpoint, the WFAT 402 and DFAT 404 are cleared by unmarking each sector listed in the WDLIST 406.
  • the DFAT 404 is cleared utilizing the WDLIST 406.
  • the WDLIST 406 is then cleared.
  • the situation immediately after the rollback of the FAT is the same as after taking a checkpoint in Figure 5 above, i.e., the filesystem is in a clean or consistent state. Therefore, the application may resume execution from the rolled back checkpoint without restarting the application.
  • Figure 7 is an exemplary method flowchart 700 that depicts a write of user data 304 to the checkpointing filesystem 400 of Figure 4 according to the present invention. It is to be noted that the steps of Figure 7 described below are behind the scenes and are not visible to the application.
  • the writing of the user data 304 to the checkpointing filesystem 400 starts at step 702.
  • sectors marked as free in the FAT 302 and not marked for deletion in the DFAT 404 are chosen for the write.
  • free sectors to contain the user data 304 that is to be written have to be found first, except that in the checkpointing filesystem 400 the sectors cannot be marked as deleted in the DFAT 404 because they are sectors marked as deleted since the previously taken checkpoint and must be preserved in case of a rollback to the previously taken checkpoint.
  • the user data 304 may be written to the free sectors at step 706.
  • the FAT 302 is updated to reflect the write of user data 304. Since the write is after a previous checkpoint, WFAT 402 and WDLIST 406 are respectively updated at steps 710 and 712.
  • the writing of user data 304 to the checkpointing filesystem 400 ends.
  • the filesystem 300 also performs a set of steps for writing the user data to the filesystem 300.
  • the conventional steps include: choosing sectors marked free in the FAT 302; writing the user data 304 to the free sectors; and updating the FAT to reflect the user data written.
  • the following steps are additional due to the checkpointing of the filesystem according to the present invention: ensuring that the free sectors are not in DFAT 404; updating WFAT 402 and updating WDLIST 406 as particularly described above.
  • the additional steps take little time as compared to the writing the user data 304.
  • checkpointing filesystem 400 of Figure 4 is almost as fast as the filesystem without checkpointing 300 of Figure 3. This fact satisfies the speed required by the distributed-memory parallel supercomputer illustrated in Figures 1 and 2.
  • Figure 8 is an exemplary method flowchart 800 describing how a delete of user data 304 from the checkpointing filesystem 400 according to the present invention.
  • the deleting of the user data 304 from the checkpointing filesystem 400 starts at step 802. It is to be noted that the steps of Figure 8 described below are behind the scenes and are not visible to the application.
  • sectors associated with user data 304 to be deleted are deleted from the FAT 302.
  • the DFAT 404 is updated at step 810 by marking in the DFAT 404 the deleted sectors. More particularly with regard to steps 806 and 808, if the sectors to be deleted are marked in the WFAT 402, then the entries are unmarked in the WFAT 402 and the DFAT 404 is not updated since the sectors deleted were created after the previous checkpoint and thus should not be recovered in a rollback. With regard to steps 806 and 810, the sectors to be deleted are not marked in the WFAT 402, and thus the sectors were created before the previous checkpoint and thus should be recovered in a rollback. Therefore, the sectors are marked in the DFAT 404.
  • the WDLIST 406 is updated at step 812 with the sectors unmarked in the WFAT 402 or marked in the DFAT 404.
  • the deleting of the user data from the checkpointing filesystem 400 ends.
  • the filesystem 300 also updates the FAT 302 with information concerning the delete.
  • the remaining steps of Figure 8 are unique to the checkpointing filesystem 400 according to the present invention. Since the remaining steps take time that is comparable to updating the FAT 302, the checkpointing filesystem according to the present invention is fast.
  • FIG. 9 is an exemplary flowchart 900 that depicts how a read of user data 304 in the checkpointing filesystem 400 according to the present invention.
  • the reading of the user data 304 from the checkpointing filesystem 400 starts at step 902.
  • the PC 202 receives a request from a node of the distributed-memory parallel supercomputer to read a file.
  • the FAT 302 is examined to determine which sectors to read.
  • the PC 202 reads the sectors determined at step 906 from the associated disk 204.
  • PC 202 transmits the read sectors to the requesting node of the distributed-memory parallel supercomputer.
  • the reading of the user data 304 from the checkpointing filesystem 400 ends.
  • the reading of user data 304 from the checkpointing filesystem is exactly as it is in a filesystem without checkpointing 300 of Figure 3. More particularly, the reads in the checkpointing filesystem 400 of distributed-memory parallel supercomputer illustrated in Figures 1 and 2 are as fast as the on the supercomputer with a filesystem without checkpointing 300.

Abstract

The present invention is directed to a checkpointing filesystem of a ditributedmemory parallel supercomputer (102) comprising a node (104) that accesses user data (110) on the filesystem (200) the filesystem (110) comprising an interface (202) that is associated with a disk (204) for storing the user data. The checkpointing filesystem provides for taking and checkpoint of the filesystem and rolling back to a previously taken checkpoint, as well as for writing user data to and deleting user data from the checkpointing filesystem. The checkpointing filesystem provides a recently written file allocation table (WFAT) for maintaining information regarding the user data written since a previously taken checkpoint and a recently deleted file allocation table (DFAT) for maintaining information regarding user data deleted from since the previously taken checkpoint, both of which are utilized by the checkpointing filesystem to take a checkpoint of the filesystem and rollback the filesystem to a previously taken checkpoint, as well as to write and delete user data from the checkpointing filesystem.

Description

CHECKPOINTING FILESYSTEM
CROSS REFERENCE
The present invention claims the benefit of commonly-owned, co-pending United States Provisional Patent Application Serial Number 60/271,124 filed February 24, 2001 entitled MASSIVELY PARALLEL SUPERCOMPUTER, the whole contents and disclosure of which is expressly incorporated by reference herein as if fully set forth herein. This patent application is additionally related to the following commonly-owned, co-pending United States Patent Applications filed on even date herewith, the entire contents and disclosure of each of which is expressly incorporated by reference herein as if fully set forth herein. U.S. patent application Serial No. (YOR920020027US1, YOR920020044US1 (15270)), for "Class Networking Routing"; U.S. patent application Serial No. (YOR920020028US1 (15271)), for "A Global Tree Network for Computing Structures"; U.S. patent application Serial No. (YOR920020029US1 (15272)), for 'Global Interrupt and Barrier Networks"; U.S. patent application Serial No. (YOR920020030US1 (15273)), for Optimized Scalable Network Switch"; U.S. patent application Serial No. (YOR920020031US1, YOR920020032US1 (15258)), for "Arithmetic Functions in Torus and Tree Networks'; U.S. patent application Serial No. (YOR920020033US1, YOR920020034US1 (15259)), for 'Data Capture Technique for High Speed Signaling"; U.S. patent application Serial No. (YOR920020035US1 (15260)), for 'Managing Coherence Via Put/Get Windows'; U.S. patent application Serial No. (YOR920020036US1, YOR920020037US1 (15261)), for "Low Latency Memory Access And Synchronization"; U.S. patent application Serial No. (YOR920020038US1 (15276), for 'Twin-Tailed Fail-Over for Fileservers Maintaining Full Performance in the Presence of Failure"; U.S. patent application Serial No. (YOR920020039US 1 (15277)), for "Fault Isolation Through No-Overhead Link Level Checksums'; U.S. patent application Serial No. (YOR920020040US1 (15278)), for "Ethernet Addressing Via Physical Location for Massively Parallel Systems"; U.S. patent application Serial No. (YOR920020041US1 (15274)), for "Fault Tolerance in a Supercomputer Through Dynamic Repartitioning"; U.S. patent application Serial No. (YOR920020042US1 (15279)), for "Checkpointing Filesystem"; U.S. patent application Serial No. (YOR920020043US1 (15262)), for "Efficient Implementation of Multidimensional Fast Fourier Transform on a Distributed-Memory Parallel Multi-Node Computer"; U.S. patent application Serial No. (YOR9-20010211US2 (15275)), for "A Novel Massively Parallel Supercomputer"; and U.S. patent application Serial No. (YOR920020045US1 (15263)), for "Smart Fan Modules and System".
BACKGROUND OF THE INVENTION
Technical Field of the Invention
The present invention generally relates to a field of fϊlesystems in computers. More particularly, the present invention is directed to efficiently checkpointing a filesystem on a distributed-memory parallel supercomputer, thereby facilitating faster execution of applications.
Description of the Prior Art
In large computing systems, such as a distributed-memory parallel supercomputer, it is standard to save a state of a system at regular intervals such that a application can be rolled back and rerun from a last saved state of the system, thereby saving time and computing resources. This is necessary because the large computing systems do not have the reliability of small computing systems and the applications that utilize the large computing systems often run for hours, days or weeks. More particularly, the large computing systems may crash or be brought down for maintenance, exceptions may be encountered while an application executes, or programmer-defined conditions may be met which terminate the application. Since the application manipulates disk files, the roll back restores the manipulated disk files to a previous clean state, i.e., to a previous checkpoint. Therefore, checkpointing the application mitigates the rerunning of the application since the application need only be rolled back to the previous checkpoint rather than be rerun from the start. Thus, checkpointing of the filesystem is a critical aspect of checkpointing in the large computing systems, such as the distributed-memory parallel supercomputer.
A main motivation for a distributed-memory parallel supercomputer is a fast execution of the application. Thus, during the execution of an application on the distributed-memory parallel supercomputer, there is a need for a checkpointing filesystem, which is not significantly slower than a filesystem without checkpointing. Similarly, there is a need for a checkpointing filesystem in which the act of checkpointing is fast, since during checkpointing of the filesystem the application is not executing. There also is a need for a checkpointing filesystem, which appears to the application as a normal filesystem and does not complicate the implementation of the application.
Therefore, there is a need in the art for providing a checkpointing filesystem on the distributed-memory parallel supercomputer that facilitates faster execution of an application executing on the distributed-memory parallel supercomputer.
SUMMARY OF THE INVENTION
It is therefore an object of the present invention to provide a checkpointing filesystem that facilitates fast execution of an application on the distributed- memory parallel supercomputer.
It is another object of the present invention to provide a checkpointing filesystem, which is not significantly slower than a filesystem without checkpointing on the distributed-memory parallel supercomputer.
It is a yet another object of the present invention to provide a checkpointing filesystem, which appears to the application as a conventional filesystem and does significantly effect an application on the distributed-memory parallel supercomputer. It is a further object of the present invention to provide a checkpointing filesystem in which the taking of a checkpoint does not significantly effect execution of an application on the distributed-memory parallel supercomputer.
It is yet a further object of the present invention to provide a checkpointing filesystem in which the rolling back of a previously taken checkpoint is efficient as possible, thereby not significantly effecting the application on the on the distributed-memory. parallel supercomputer.
According to an embodiment of the present invention, there is provided a method for checkpointing a filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: informing the node and the interface to complete their access to the filesystem in order to take a checkpoint of the filesystem; directing the interface to take the checkpoint of the associated disk, the taking of the checkpoint comprising a step of clearing: (i) a recently written file allocation table (WFAT) for maintaining information regarding the user data written to the disk since a previously taken checkpoint; (ii) a recently deleted file allocation table (DFAT) for maintaining information regarding the user data deleted from the disk since the previously taken checkpoint; and informing the node and the interface that they may resume their access to the filesystem, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
According to another embodiment of the present invention, there is provided a method for rolling back a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem to a previously taken checkpoint of the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: informing the node and the interface to complete their access to the filesystem in order rollback to the previously taken checkpoint of the filesystem; directing the interface to rollback to the previously taken checkpoint of the associated disk, the rolling back to the previously taken checkpoint comprising: (i) removing user data from a file allocation table (FAT) utilizing a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since the previously taken checkpoint; (ii) adding user data to the FAT from a recently deleted file allocation table (DFAT) that maintains information regarding user data deleted from the disk since the previously taken checkpoint; and informing the node and the interface that they may resume their access to the filesystem, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
According to yet another embodiment of the present invention, there is provided a method for writing user data to a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses the user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: selecting sectors that are marked as free in a file allocation table (FAT) and not marked as deleted in a recently deleted file allocation table (DFAT) that maintains information regarding user data deleted from the disk since a previously taken checkpoint; writing the user data to the selected sectors on the disk; updating the FAT to reflect the written user data; and updating a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
According to a further embodiment of the present invention, there is provided a method for deleting user data from a checkpointing filesystem of a distributed- memory parallel supercomputer comprising a node that accesses the user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: deleting sectors associated with the user data to be deleted from the disk from a file allocation table (FAT); determining whether the sectors to be deleted are marked in a recently written file allocation table (WFAT) that maintains information regarding user data written to the disk since a previously taken checkpoint; and updating a recently deleted file allocation table (DFAT) if the sectors are not marked in WFAT at step (b), the DFAT maintaining information regarding the user data deleted from the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
According to yet a further embodiment, there is provided a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, comprising: a host controller comprising means for informing the node and the interface to complete their access to the filesystem in order to take a checkpoint of the filesystem, directing the interface to take a checkpoint of the associated disk, and informing the node and the interface that they may resume their access to the filesystem after the checkpoint it taken; and the interface comprising a means for clearing: a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since a previously taken checkpoint; and a recently deleted file allocation table (DFAT) for maintaining information regarding the user data deleted from the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
The checkpointing filesystem of the distributed-memory parallel supercomputer further comprises rolling back to the previously taken checkpoint, wherein: the host controller further comprises means for informing the node and the interface to complete their access to the filesystem in order rollback to the previously taken checkpoint of the filesystem, for directing the interface to rollback to the previously taken checkpoint of the associated disk, and for informing the node and the interface that they may resume their access to the filesystem; and the interface further comprises means for rolling back to the previously taken checkpoint, the means comprising removing user data from a file allocation table (FAT) utilizing the WFAT and adding user data to the FAT from DFAT.
The checkpointing filesystem of the distributed-memory parallel supercomputer further comprising writing user data to the checkpointing filesystem, wherein: the interface further comprises means for selecting sectors that are marked as free in the FAT and not marked for deletion in DFAT, for writing the user data to the selected sectors on the disk; for updating the FAT to reflect the written user data, and for updating WFAT.
The checkpointing filesystem of a distributed-memory parallel supercomputer further comprising deleting user data from the checkpointing filesystem, wherein: the interface further comprises means for deleting sectors associated with the user data to be deleted from the disk from the FAT, for determining whether the sectors to be deleted are marked in the WFAT, and for updating the DFAT if the sectors are not marked in WFAT.
BRIEF DESCRIPTION OF THE DRAWINGS
The objects, features and advantages of the present invention will become apparent to one skilled in the art, in view of the following detailed description taken in combination with the attached drawings, in which:
Figure 1 illustrates an exemplary high-level system diagram for performing checkpointing of the filesystem according to the present invention;
Figure 2 illustrates is an exemplary more-detailed system diagram of Figure 1 for performing checkpointing of the filesystem according to the present invention;
Figure 3 illustrates a conventional filesystem without checkpointing;
Figure 4 illustrates an exemplary checkpointing filesystem according to the present invention;
Figure 5 illustrates an exemplary method flowchart that depicts taking a checkpoint in the checkpointing filesystem of Figure 4 according to the present invention; Figure 6 illustrates an exemplary method flowchart that depicts rolling back to a previously taken checkpoint in the checkpointing filesystem of Figure 4 according to the present invention;
Figure 7 illustrates an exemplary method flowchart that depicts writing of user data to the checkpointing filesystem 400 of Figure 4 according to the present invention;
Figure 8 illustrates an exemplary method flowchart that depicts deleting of user data from the checkpointing filesystem 400 according to the present invention; and
Figure 9 illustrates an exemplary flowchart that depicts reading of user data in the checkpointing filesystem 400 according to the present invention.
DETAILED DESCRIPTION OF THE
PREFERRED EMBODIMENT OF THE INVENTION
The present invention is directed a checkpointing filesystem on the distributed- memory parallel supercomputer that facilitates faster execution of an application executing on the distributed-memory parallel supercomputer.
Figure 1 is an exemplary high-level system diagram 100 for performing checkpointing of the filesystem according to the present invention. The high performance computer engine (i.e., "HPCE") 102 represents the distributed- memory parallel supercomputer. The HPCE is a distributed-memory parallel supercomputer that comprises a plurality of nodes. The HPCE 102 communicates via the network switch 104 with the fileserver/filesystem 106 and host control server 108. The HPCE communicates with the foregoing system components over communication connections 110, which may represent standard networking connections, such as Ethernet and the like. The host control server 108 controls the HPCE 102 and the fileserver/filesystem 106, and determines when to perform a checkpoint of the filesystem and when to restore the filesystem from a previous checkpoint as will be described below in greater detail. Figure 2 is an exemplary more-detailed system diagram 200 of Figure 1 for performing checkpointing of the filesystem according to the present invention. The fileserver/filesystem 106 comprises a plurality of personal computers (i.e., "PCs") 202. The distributed-memory parallel supercomputer comprises approximately 1024 PCs. In the distributed-memory parallel supercomputer, each PC 202 serves a Redundant Array of Independent Disks (i.e., "RAID") cage, i.e., a plurality of disk drives. For brevity and clarity of the present invention, the following description will be in terms of each PC 202 serving a disk 204 instead of a RAID cage. More particularly, each PC 202 serves as an interface for the HPCE 102 to an associated disk 204 via the network switch 104. The HPCE 102 communicates with the fileserver/filesystem 106 via messages that represent basic file operations, such as read, write and delete. A node of the HPCE 102 may access any file stored on any disk 204 via associated PC interface 202.
Further with reference to Figure 2, in order to perform a checkpoint, the host control 108 informs each of the nodes of the HPCE 102 and the PCs 202 to complete any disk activity (i.e., access) and not to start any new disk activity. The host control 108 directs each PC 202 to make a checkpoint of its associated disk 204. This checkpointing activity is completely local to each PC 202 and its associated disk 204. The checkpointing activity is the same with regard to all PCs, i.e., as if a single PC 202 performs a checkpoint of its associated disk 204.
Similarly, when a checkpoint is restored, the restoration is completely local to each PC 202 and its associated disk 204. Therefore, in the following description performing a checkpoint and restoring a checkpoint will be described in detail with regard to a single PC 202 and associated drive 204. The host control 108 is a complete computer including its own internal filesystem, which maintains the software for driving the host control 108, including the taking of a checkpoint and rolling back to previous checkpoint as described below in greater detail. In fileserver/filesystem 106, each PC 202 is a complete computer including its own internal filesystem, in addition to the associated disk 204. The internal filesystem of the PC 202 maintains the software for driving the PC 202, including the taking of a. checkpoint and rolling back to a previous checkpoint as described below in greater detail. Figure 3 depicts a conventional filesystem 300 without checkpointing. The filesystem 300 occupies an entire disk or a portion of the disk. The filesystem comprises two parts. The first part is user data 304 and the free space 306 in the filesystem. The second part is a File Allocation Table (i.e., "FAT"), also known as metadata, which includes a description of the user data 304 and the free space 306. Thus, the FAT 302 includes filenames, file data start sectors, file sizes, as well as free sectors. An example of such a FAT 302 is the industry standard FAT32.
Figure 4 is an exemplary checkpointing filesystem 400 according to the present invention. The filesystem comprises five parts. The first part is the user data 304 and the free space 306 in the checkpointing filesystem 400, which are analogous to the filesystem without checkpointing 300 in Figure 3. The second part is the File Allocation Table (i.e., "FAT") 302, which is analogous to that of the filesystem without checkpointing 300 in Figure 3. The third part of the checkpointing filesystem 400 is a recently Written FAT (i.e., "WFAT") 402, which is structured like the FAT 302, but only maintains information on the user data 304 written since a previous checkpoint. The fourth part of the checkpointing filesystem 400 is a recently Deleted FAT (i.e., "DFAT") 404, which is structured like the FAT 302, but only maintains information on the user data 304 deleted since the previous checkpoint. The fifth and final part the checkpointing filesystem 400 is the recently Written or Deleted list (i.e., "WDLIST") of user data, which includes the same information as the WFAT 402 and the DFAT 404. The WFAT 402 and DFAT 404 are optimal for fast random access, since they provide indexed by sector access to information about recently written or deleted user data 304. Like the FAT 302, the WFAT 402 includes an entry for every sector in the checkpointing filesystem 400. Typically only a small fraction, if any, of sectors includes user data 304 written since the previous checkpoint, so most of the WFAT 402 entries are empty. Similarly, most of the DFAT 404 entries are empty. By contrast to the WFAT 402 and DFAT 404, the WDLIST 406 is optimal for fast sequential access to all the information about the recently written or deleted user data 304. More particularly, the WDLIST 406 provides a list of sectors in the WFAT 402 marking data written since the previous checkpoint and of sectors in the DFAT 404 marking data deleted since the previous checkpoint. Figure 5 is an exemplary method flowchart 500 that depicts taking a checkpoint in the checkpointing filesystem 400 according to the present invention. The taking of the checkpoint starts at step 502. At step 504, the WFAT 402 is cleared, since at the checkpoint the user data 304 written since a previous checkpoint becomes part of the checkpoint. Since the WFAT 402, just like the FAT 302, is a large data structure, clearing the WFAT 402 by iterating through the entire data structure is inefficient. Because the WDLIST 406 is a much smaller data structure than the WFAT 402, it is faster to clear the WFAT 402 using entries in the WDLIST 406. At step 506, the DFAT 404 is cleared since at the checkpoint the user data 304 deleted since the previous checkpoint must be deleted from the checkpoint. For efficiency, the DFAT 404 is cleared using WDLIST 406. More particularly, the WDLIST 406 is a list of all sectors in the WFAT 402 and DFAT 404 marking user data 304 written or deleted since the previous checkpoint. The WFAT 404 and DFAT 404 are cleared by unmarking each sector listed in the WDLIST 406. After the WFAT 402 and the DFAT 404 are respectively cleared at steps 504 and 506, the WDLIST 406 is also cleared by deleting its list of marked sectors. At this point the filesystem is in a clean or consistent state. The exemplary taking of the checkpoint 500 is fast as required by the distributed-memory parallel supercomputer illustrated in Figures 1 and 2, and minimally impacts the application.
Figure 6 is an exemplary method flowchart 600 that depicts rolling back to a previously taken checkpoint according to the present invention. The rolling back to the previously taken checkpoint starts at step 602. At step 604, the FAT 302 is returned to the previously taken checkpoint using WDLIST 406. More particularly, at the rollback to the previously taken checkpoint, the user data 304 written since the previous checkpoint is removed from the FAT 302, and the user data 304 deleted since the previous checkpoint is returned to the FAT 302. Since the WDLIST 406 includes all the information on the user data 304 written and deleted since the previously taken checkpoint, the WDLIST is used to return the FAT to the previously taken checkpoint. At step 606, the WFAT 402 is cleared utilizing WDLIST 406 for efficiency. Since, the WDLIST 406 is a list of all sectors in the WFAT 402 and DFAT 404 marking user data 304 written or deleted since the previous checkpoint, the WFAT 402 and DFAT 404 are cleared by unmarking each sector listed in the WDLIST 406. At step 608, the DFAT 404 is cleared utilizing the WDLIST 406. At step 610, the WDLIST 406 is then cleared. The situation immediately after the rollback of the FAT is the same as after taking a checkpoint in Figure 5 above, i.e., the filesystem is in a clean or consistent state. Therefore, the application may resume execution from the rolled back checkpoint without restarting the application.
Figure 7 is an exemplary method flowchart 700 that depicts a write of user data 304 to the checkpointing filesystem 400 of Figure 4 according to the present invention. It is to be noted that the steps of Figure 7 described below are behind the scenes and are not visible to the application. The writing of the user data 304 to the checkpointing filesystem 400 starts at step 702. At step 704, sectors marked as free in the FAT 302 and not marked for deletion in the DFAT 404 are chosen for the write. More particularly, as in the case of the filesystem without checkpointing 300 of Figure 3, free sectors to contain the user data 304 that is to be written have to be found first, except that in the checkpointing filesystem 400 the sectors cannot be marked as deleted in the DFAT 404 because they are sectors marked as deleted since the previously taken checkpoint and must be preserved in case of a rollback to the previously taken checkpoint. Given that the free sectors are not in the DFAT 402, the user data 304 may be written to the free sectors at step 706. At step 708, the FAT 302 is updated to reflect the write of user data 304. Since the write is after a previous checkpoint, WFAT 402 and WDLIST 406 are respectively updated at steps 710 and 712. At step 714, the writing of user data 304 to the checkpointing filesystem 400 ends. In the conventional filesystem 300 without checkpointing, the filesystem 300 also performs a set of steps for writing the user data to the filesystem 300. The conventional steps include: choosing sectors marked free in the FAT 302; writing the user data 304 to the free sectors; and updating the FAT to reflect the user data written. Over and above the conventional filesystem 300, the following steps are additional due to the checkpointing of the filesystem according to the present invention: ensuring that the free sectors are not in DFAT 404; updating WFAT 402 and updating WDLIST 406 as particularly described above. The additional steps take little time as compared to the writing the user data 304. For example, ensuring that the sectors are not in DFAT is a fast random access lookup. Thus the checkpointing filesystem 400 of Figure 4 according to the present invention is almost as fast as the filesystem without checkpointing 300 of Figure 3. This fact satisfies the speed required by the distributed-memory parallel supercomputer illustrated in Figures 1 and 2.
Figure 8 is an exemplary method flowchart 800 describing how a delete of user data 304 from the checkpointing filesystem 400 according to the present invention. The deleting of the user data 304 from the checkpointing filesystem 400 starts at step 802. It is to be noted that the steps of Figure 8 described below are behind the scenes and are not visible to the application. At step 804, sectors associated with user data 304 to be deleted are deleted from the FAT 302. At step 806, it is determined whether the sectors to be deleted from the FAT 302 are marked in the WFAT 402. If the sectors to be deleted are marked in the WFAT 402, then the sectors are deleted from the WFAT 402 at step 808. Otherwise, if the sectors to be deleted are not marked in the WFAT 402, the DFAT 404 is updated at step 810 by marking in the DFAT 404 the deleted sectors. More particularly with regard to steps 806 and 808, if the sectors to be deleted are marked in the WFAT 402, then the entries are unmarked in the WFAT 402 and the DFAT 404 is not updated since the sectors deleted were created after the previous checkpoint and thus should not be recovered in a rollback. With regard to steps 806 and 810, the sectors to be deleted are not marked in the WFAT 402, and thus the sectors were created before the previous checkpoint and thus should be recovered in a rollback. Therefore, the sectors are marked in the DFAT 404. In either case, whether or not the sectors to be deleted are marked in the WFAT 402, the WDLIST 406 is updated at step 812 with the sectors unmarked in the WFAT 402 or marked in the DFAT 404. At step 814, the deleting of the user data from the checkpointing filesystem 400 ends. In the conventional filesystem 300 without checkpointing, the filesystem 300 also updates the FAT 302 with information concerning the delete. However, the remaining steps of Figure 8 are unique to the checkpointing filesystem 400 according to the present invention. Since the remaining steps take time that is comparable to updating the FAT 302, the checkpointing filesystem according to the present invention is fast.
Figure 9 is an exemplary flowchart 900 that depicts how a read of user data 304 in the checkpointing filesystem 400 according to the present invention. The reading of the user data 304 from the checkpointing filesystem 400 starts at step 902. At step 904, the PC 202 receives a request from a node of the distributed-memory parallel supercomputer to read a file. At step 906, the FAT 302 is examined to determine which sectors to read. At step 908, the PC 202 reads the sectors determined at step 906 from the associated disk 204. At step 910, PC 202 transmits the read sectors to the requesting node of the distributed-memory parallel supercomputer. At step 912, the reading of the user data 304 from the checkpointing filesystem 400 ends. Thus, the reading of user data 304 from the checkpointing filesystem is exactly as it is in a filesystem without checkpointing 300 of Figure 3. More particularly, the reads in the checkpointing filesystem 400 of distributed-memory parallel supercomputer illustrated in Figures 1 and 2 are as fast as the on the supercomputer with a filesystem without checkpointing 300.
While the invention has been particularly shown and described with regard to preferred embodiments thereof, it will be understood by those skilled in the art that the foregoing and other changes in form and details may be made therein without d eparting from the spirit and scope of the invention.

Claims

CLAIMS:Having thus described our invention, what we claim as new, and desire to secure by Letters Patent is:
1. A method for checkpointing a filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: (a) informing the node and the interface to complete their access to the filesystem in order to take a checkpoint of the filesystem; (b) directing the interface to take the checkpoint of the associated disk, the taking of the checkpoint comprising a step of clearing: (i) a recently written file allocation table (WFAT) for maintaining information regarding the user data written to the disk since a previously taken checkpoint; (ii) a recently deleted file allocation table (DFAT) for maintaining information regarding the user data deleted from the disk since the previously taken checkpoint; (c) informing the node and the interface that they may resume their access to the filesystem, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
2. The method for checkpointing a filesystem of a distributed-memory parallel supercomputer according to Claim 1, wherein the WFAT and the DFAT are accessible in random order.
3. The method for checkpointing a filesystem of a distributed-memory parallel supercomputer according to Claim 1, wherein a written or deleted list (WDLIST) is provided for maintaining information associated with the WFAT and the DFAT and is accessible in a sequential order.
4. The method for checkpointing a filesystem of a distributed-memory parallel supercomputer according to according to Claim 3, wherein the step of clearing utilizes the information of the WDLIST to clear the WFAT and DFAT.
5. The method for checkpointing a filesystem of a distributed-memory parallel supercomputer according to according to Claim 4, wherein the method further comprises a step of clearing the WDLIST.
6. A method for rolling back a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem to a previously taken checkpoint of the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: (a) informing the node and the interface to complete their access to the filesystem in order rollback to the previously taken checkpoint of the filesystem; (b) directing the interface to rollback to the previously taken checkpoint of the associated disk, the rolling back to the previously taken checkpoint comprising: (i) removing user data from a file allocation table (FAT) utilizing a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since the previously taken checkpoint; (ii) adding user data to the FAT from a recently deleted file allocation table (DFAT) that maintains information regarding user data deleted from the disk since the previously taken checkpoint; and (c) informing the node and the interface that they may resume their access to the filesystem, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
7. The method for rolling back a checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 6, wherein the WFAT and the DFAT are accessible in random order.
8. The method for rolling back a checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 1 , wherein a written or deleted list (WDLIST) is provided for maintaining information associated with the WFAT and the DFAT and is accessible in a sequential order.
9. The method for rolling back a checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 8, wherein the steps of removing and adding user data utilize the information of the WDLIST.
10. The method for rolling back a checkpointing filesystem of a distributed- memory parallel supercomputer according to Claim 9, wherein the method further comprises the steps of: clearing the WFAT and the DFAT utilizing information of the WDLIST; and clearing the WDLIST.
11. A method for writing user data to a checkpointing filesystem of a distributed- memory parallel supercomputer comprising a node that accesses the user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: (a) selecting sectors that are marked as free in a file allocation table (FAT) and not marked as deleted in a recently deleted file allocation table (DFAT) that maintains information regarding user data deleted from the disk since a previously taken checkpoint; (b) writing the user data to the selected sectors on the disk - (c) updating the FAT to reflect the written user data; and (d) updating a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since the previously taken checkpoint; wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
12. The method for writing user data to a checkpointing filesystem of a distributed- memory parallel supercomputer according to Claim 11 , wherein the WFAT and the DFAT are accessible in random order.
13. A method for writing user data to a checkpointing filesystem of a distributed- memory parallel supercomputer Claim 11 , wherein a written or deleted list (WDLIST) is provided for maintaining information associated with the WFAT and the DFAT and is accessible in a sequential order.
14. A method for deleting user data from a checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses the user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, the method comprising the steps of: (a) deleting sectors associated with the user data to be deleted from the disk from a file allocation table (FAT); (b) determining whether the sectors to be deleted are marked in a recently written file allocation table (WFAT) that maintains information regarding user data written to the disk since a previously taken checkpoint; and (c) updating a recently deleted file allocation table (DFAT) if the sectors are not marked in WFAT at step (b), the DFAT maintaining information regarding the user data deleted from the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
15. The method for deleting user data from a checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 14, wherein the WFAT and the DFAT are accessible in random order.
16. The method for deleting user data from a checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 14, wherein a written or deleted list (WDLIST) is provided for maintaining information associated with the WFAT and the DFAT and is accessible in a sequential order.
17. The method for deleting user data from a checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 16, further comprising a step of updating the WDLIST.
18. A checkpointing filesystem of a distributed-memory parallel supercomputer comprising a node that accesses user data on the filesystem, the filesystem comprising an interface that is associated with a disk for storing the user data, comprising: a host controller comprising means for informing the node and the interface to complete their access to the filesystem in order to take a checkpoint of the filesystem, directing the interface to take a checkpoint of the associated disk, and informing the node and the interface that they may resume their access to the filesystem after the checkpoint it taken; and the interface comprising a means for clearing: a recently written file allocation table (WFAT) that maintains information regarding the user data written to the disk since a previously taken checkpoint; and a recently deleted file allocation table (DFAT) for maintaining information regarding the user data deleted from the disk since the previously taken checkpoint, wherein the WFAT and DFAT are utilized to rollback the filesystem to a clean state.
19. The checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 18, further comprising rolling back to the previously taken checkpoint, wherein: the host controller further comprising means for informing the node and the interface to complete their access to the filesystem in order rollback to the previously taken checkpoint of the filesystem, for directing the interface to rollback to the previously taken checkpoint of the associated disk, and for informing the node and the interface that they may resume their access to the filesystem; and the interface further comprising means for rolling back to the previously taken checkpoint, the means comprising: removing user data from a file allocation table (FAT) utilizing the WFAT and adding user data to the FAT from DFAT.
20. The checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 18, further comprising writing user data to the checkpointing filesystem: the interface further comprising means for selecting sectors that are marked as free in the FAT and not marked for deletion in DFAT, for writing the user data to the selected sectors on the disk; for updating the FAT to reflect the written user data, and for updating WFAT.
21. The checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 18, further comprising deleting user data from the checkpointing filesystem: the interface further comprising means for deleting sectors associated with the user data to be deleted from the disk from the FAT, for determining whether the sectors to be deleted are marked in the WFAT, and for updating the DFAT if the sectors are not marked in WFAT.
22. The checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 18, wherein the WFAT and the DFAT are accessible in random order.
23. The checkpointing filesystem of a distributed-memory parallel supercomputer according to Claim 18, wherein a written or deleted list (WDLIST) is provided for maintaining information associated with the WFAT and the DFAT and is accessible in a sequential order.
PCT/US2002/005613 2001-02-24 2002-02-25 Checkpointing filesystem WO2002069200A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/US2002/005613 WO2002069200A1 (en) 2001-02-24 2002-02-25 Checkpointing filesystem
US10/258,515 US6895416B2 (en) 2001-02-24 2002-02-25 Checkpointing filesystem

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
US27112401P 2001-02-24 2001-02-24
US60/271,124 2001-02-24
PCT/US2002/005613 WO2002069200A1 (en) 2001-02-24 2002-02-25 Checkpointing filesystem

Publications (1)

Publication Number Publication Date
WO2002069200A1 true WO2002069200A1 (en) 2002-09-06

Family

ID=68499857

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2002/005613 WO2002069200A1 (en) 2001-02-24 2002-02-25 Checkpointing filesystem

Country Status (2)

Country Link
US (1) US6895416B2 (en)
WO (1) WO2002069200A1 (en)

Families Citing this family (32)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1322452C (en) * 2001-02-24 2007-06-20 国际商业机器公司 Arithmetric functions in torus and tree networks
US6643654B1 (en) * 2001-06-25 2003-11-04 Network Appliance, Inc. System and method for representing named data streams within an on-disk structure of a file system
US7139781B2 (en) * 2003-04-29 2006-11-21 International Business Machines Corporation Managing filesystem versions
US7478275B1 (en) * 2004-03-29 2009-01-13 Symantec Operating Corporation Method and apparatus for performing backup storage of checkpoint data within a server cluster
JP4561168B2 (en) * 2004-04-28 2010-10-13 株式会社日立製作所 Data processing system and method, and processing program therefor
US7478278B2 (en) * 2005-04-14 2009-01-13 International Business Machines Corporation Template based parallel checkpointing in a massively parallel computer system
JP4927408B2 (en) 2006-01-25 2012-05-09 株式会社日立製作所 Storage system and data restoration method thereof
US8533778B1 (en) * 2006-06-23 2013-09-10 Mcafee, Inc. System, method and computer program product for detecting unwanted effects utilizing a virtual machine
JP5595633B2 (en) * 2007-02-26 2014-09-24 スパンション エルエルシー Simulation method and simulation apparatus
US7870122B2 (en) * 2007-04-23 2011-01-11 Microsoft Corporation Self-tuning index for flash-based databases
US8037112B2 (en) * 2007-04-23 2011-10-11 Microsoft Corporation Efficient access of flash databases
JP2009104300A (en) * 2007-10-22 2009-05-14 Denso Corp Data processing apparatus and program
US7962780B2 (en) * 2008-08-22 2011-06-14 International Business Machines Corporation Command sequence numbering apparatus and method
US8458403B2 (en) * 2009-11-24 2013-06-04 Honeywell International Inc. Architecture and method for cache-based checkpointing and rollback
US8108721B2 (en) * 2009-11-24 2012-01-31 Honeywell International Inc. Architecture and method for hardware-assisted processor checkpointing and rollback
US8402064B2 (en) * 2010-02-01 2013-03-19 Oracle International Corporation Orchestration of business processes using templates
US8307243B2 (en) * 2010-02-01 2012-11-06 International Business Machines Corporation Parallel debugging in a massively parallel computing system
US20110218921A1 (en) * 2010-03-05 2011-09-08 Oracle International Corporation Notify/inquire fulfillment systems before processing change requests for adjusting long running order management fulfillment processes in a distributed order orchestration system
US10061464B2 (en) * 2010-03-05 2018-08-28 Oracle International Corporation Distributed order orchestration system with rollback checkpoints for adjusting long running order management fulfillment processes
US20110218923A1 (en) * 2010-03-05 2011-09-08 Oracle International Corporation Task layer service patterns for adjusting long running order management fulfillment processes for a distributed order orchestration system
US20110218925A1 (en) * 2010-03-05 2011-09-08 Oracle International Corporation Change management framework in distributed order orchestration system
US10789562B2 (en) * 2010-03-05 2020-09-29 Oracle International Corporation Compensation patterns for adjusting long running order management fulfillment processes in an distributed order orchestration system
US10395205B2 (en) * 2010-03-05 2019-08-27 Oracle International Corporation Cost of change for adjusting long running order management fulfillment processes for a distributed order orchestration system
US9904898B2 (en) * 2010-03-05 2018-02-27 Oracle International Corporation Distributed order orchestration system with rules engine
US20110218926A1 (en) * 2010-03-05 2011-09-08 Oracle International Corporation Saving order process state for adjusting long running order management fulfillment processes in a distributed order orchestration system
US9269075B2 (en) * 2010-03-05 2016-02-23 Oracle International Corporation Distributed order orchestration system for adjusting long running order management fulfillment processes with delta attributes
US8793262B2 (en) * 2010-03-05 2014-07-29 Oracle International Corporation Correlating and mapping original orders with new orders for adjusting long running order management fulfillment processes
US9658901B2 (en) 2010-11-12 2017-05-23 Oracle International Corporation Event-based orchestration in distributed order orchestration system
US10552769B2 (en) 2012-01-27 2020-02-04 Oracle International Corporation Status management framework in a distributed order orchestration system
US8762322B2 (en) 2012-05-22 2014-06-24 Oracle International Corporation Distributed order orchestration system with extensible flex field support
US9672560B2 (en) 2012-06-28 2017-06-06 Oracle International Corporation Distributed order orchestration system that transforms sales products to fulfillment products
CN107885598A (en) * 2017-09-26 2018-04-06 东软集团股份有限公司 Storage device detection method, device and computer

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5551043A (en) * 1994-09-07 1996-08-27 International Business Machines Corporation Standby checkpoint to prevent data loss
US5864849A (en) * 1996-12-16 1999-01-26 Lucent Technologies Inc. System and method for restoring a multiple checkpointed database in view of loss of volatile memory
US6173292B1 (en) * 1998-03-04 2001-01-09 International Business Machines Corporation Data recovery in a transactional database using write-ahead logging and file caching

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6058400A (en) * 1998-04-28 2000-05-02 Sun Microsystems, Inc. Highly available cluster coherent filesystem

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5551043A (en) * 1994-09-07 1996-08-27 International Business Machines Corporation Standby checkpoint to prevent data loss
US5864849A (en) * 1996-12-16 1999-01-26 Lucent Technologies Inc. System and method for restoring a multiple checkpointed database in view of loss of volatile memory
US6173292B1 (en) * 1998-03-04 2001-01-09 International Business Machines Corporation Data recovery in a transactional database using write-ahead logging and file caching

Also Published As

Publication number Publication date
US6895416B2 (en) 2005-05-17
US20030078933A1 (en) 2003-04-24

Similar Documents

Publication Publication Date Title
US6895416B2 (en) Checkpointing filesystem
US9727430B2 (en) Failure recovery method in information processing system and information processing system
US8074035B1 (en) System and method for using multivolume snapshots for online data backup
CN101377750B (en) System and method for cluster fault toleration
EP1918836B1 (en) Apparatus and method for a hardware-based file system
US8239356B2 (en) Methods and apparatuses for data protection
US7885938B1 (en) Techniques for granular recovery of data from local and remote storage
US7281104B1 (en) System and method for online data migration
US7406487B1 (en) Method and system for performing periodic replication using a log
CN102792276B (en) Buffer disk in flashcopy cascade
JP4620457B2 (en) Multiple simultaneously active file systems
JP4839091B2 (en) Database recovery method and computer system
US20060174075A1 (en) Method for creating and preserving snapshots in a storage system
CN107656834B (en) System and method for recovering host access based on transaction log and storage medium
GB2369206A (en) Excluding last written segments while rebuilding meta-data in a data storage system
US10261696B2 (en) Performance during playback of logged data storage operations
US7165160B2 (en) Computing system with memory mirroring and snapshot reliability
US7398420B2 (en) Method for keeping snapshot image in a storage system
US10656867B2 (en) Computer system, data management method, and data management program
CN114253765A (en) Data recovery method and related equipment
US20140297945A1 (en) Storage apparatus and copy control method
JPH10254787A (en) Guarantee processor for equivalence of duplex volume
JPH09305469A (en) File exclusive control system

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 10258515

Country of ref document: US

121 Ep: the epo has been informed by wipo that ep was designated in this application
REG Reference to national code

Ref country code: DE

Ref legal event code: 8642

122 Ep: pct application non-entry in european phase
NENP Non-entry into the national phase

Ref country code: JP

WWW Wipo information: withdrawn in national office

Country of ref document: JP