WO2004032429A1

WO2004032429A1 - Access link bandwidth management scheme

Info

Publication number: WO2004032429A1
Application number: PCT/EP2002/011005
Authority: WO
Inventors: Marcus Alexander Samsonovits; Rolf Georg Engstrand
Original assignee: Telefonaktiebolaget Lm Ericsson (Publ)
Priority date: 2002-10-01
Filing date: 2002-10-01
Publication date: 2004-04-15
Also published as: AU2002340961A1

Abstract

A scheme for allocating transmission bandwidth for access to a broadband packet-switched network, such as the Internet, to network users is proposed. This is achieved by providing at least two channels in the access link with each channel having a reserved proportion of the total transmission bandwidth. The channels may be real physical network connections or time slots that may be multiplexed onto fewer physical network connections. The information exchange to and from each user is monitored, and, as a function of the monitored information exchange rate, the users are divided into groups with users with a high information exchange being in the same group. Each group is then assigned to a channel in the access link so that all information exchange for users of a particular group is routed through the assigned channel. Each user in a group then has equal access to the allocated proportion of bandwidth.

Description

Access link bandwidth management scheme

Field of invention The invention relates to the management of bandwidth of an access link, and specifically of an access link to a packet-switched access network such as the Internet.

Background art A pricing scheme commonly offered by Internet service providers (ISP) to their customers allows all users equal access to transmission resources up to a maximum information rate for a fixed cost per month. The fixed cost option is valued by customers and the ISPs alike as it imposes a fixed budget on Internet usage on the one hand and simplifies billing on the other. Under this scheme some customers will inevitably extract more value from the service than others. For example, some users may regularly download large files and also browse or surf the Internet intensively and often, while others connect to the Internet only now and then to surf or check emails. While this may seem reasonable under the terms of the scheme, the limited bandwidth shared between any group of users means that the consistently high rate of information transfer demanded by some users puts those users that have a low or intermittent high rate of information transfer at a disadvantage. Those users that connect to the Internet only occasionally will expect a fast connection most of the time to feel they are getting value for money. When bandwidth is shared with more intensive users, the occasional users are likely to perceive a slower connection and thus a lower quality of service than they expect.

US-A-6 085 241 describes a bandwidth manager for controlling access to

Internet bandwidth by a group of users. The bandwidth manager computes bandwidth usage statistics packet-by-packet for each IP-address that is managed. The arrangement permits each individual IP-address to be assigned a specific bandwidth. Bandwidth excesses are actively controlled by delaying or dropping packets to or from the specific IP customer or address that has exceeded its bandwidth limitation. This arrangement requires a significant hardware overhead to continuously monitor and actively control the traffic from all managed IP addresses.

In view of this prior art it is an object of the present invention to improve the perceived quality of service for different types of user in a simple yet flexible manner.

It is a further object of the present invention to provide a method and arrangement for improving the perceived quality of service of Internet users without the need for specific control of excess bandwidth usage.

Summary of the invention

These and further objects of the invention are obtained in a method and arrangement for allocating transmission bandwidth to network users in accordance with the claims.

In accordance with the invention this allocation of bandwidth is achieved by providing at least two channels in the access link with each channel having a defined reserved proportion of the total transmission bandwidth. This proportion may be fixed or variable. The channels may be real physical network connections, or virtual channels in the form of time slots that may be multiplexed onto fewer physical network connections. The information exchange to and from each user is then monitored, and, as a function of the monitored information exchange rate, the users are divided into groups such that users with a similar rate of information exchange rate are in the same group. Each group is then assigned to a channel in the access link so that all information exchange for users of a particular group are routed through the assigned channel. Each user in a group then has equal access only to the allocated proportion of bandwidth.

In this manner, users that have a similar average rate of information exchange can be grouped together with exclusive shared access to a defined bandwidth that is less than the total bandwidth available. The inequality of prior art schemes due to mixing of users having different bandwidth usage practices is thus obviated. Moreover, the sorting mechanism in accordance with the present invention is highly adaptive without the need for active control as each user is assigned a shared proportion of bandwidth solely on the basis of that user's usage characteristics. The rate of access to that shared bandwidth need depend only on the number of active users in the group at any one time.

BRIEF DESCRIPTION OF THE DRAWINGS

Further objects and advantages of the present invention will become apparent from the following description of the preferred embodiments that are given by way of example with reference to the accompanying drawings. In the figures:

Fig. 1 schematically illustrates the function of a resource arbiter for bandwidth allocation in accordance with the present invention;

Fig. 2 schematically depicts the functional elements of a resource arbiter in accordance with the present invention;

Fig. 3 illustrates an alternative embodiment of a resource arbiter in accordance with the present invention, and

Fig. 4 is a flow diagram schematically illustrating one method for allocating users to groups. DETAILED DESCRIPTION OF THE DRAWINGS

Fig. 1 schematically shows a resource arbiter 10 for connecting a plurality of user ports 20 through a number of network connections or uplinks 30 to a network (not shown). The term 'user' and 'user port 20' are employed interchangeably in this document to indicate both individual users with individual network or IP addresses as well as groups of users, for example, collected under a group IP address.

The user ports 20 are preferably connected to the resource arbiter 10 through a broad band port (not illustrated). The user ports 20 may include a range of different connections, including, but not limited to, Fast Ethernet, ADSL or cable modem. The network connections 30 together constitute an access link 310 to a broadband packet-switched network, such as the Internet. The network connections 30 may be any suitable connections, an example being

Gbit Ethernet ports. Both the user ports 20 and the network connections 30 are bidirectional as indicated by the double headed arrows. The resource arbiter 10 thus controls the routing of traffic in both directions. The connection is scaled such that the sum of the bandwidths of the user ports 20 is much higher than the total bandwidth provided by the access link 310. This ensures that the bandwidth capacity of the access link 310 is used efficiently, but also brings about the risk of congestion.

If all user ports 20 in this arrangement are given equal access to the access link 310, for example using round robin scheduling or weighted fair queuing, a minority of users 20 could appropriate a disproportionate amount of the total bandwidth provided by the access links 30. This could, for example, be achieved by this minority of users receiving or sending large amounts of data over extended periods of time, such as by downloading large files regularly. Any user attempting to access the network during these periods but requiring only intermittent but high rates of data transfer, for example when surfing the Internet, will naturally be limited to a lower rate of transfer and will experience higher levels of latency than is comfortable or tolerable.

In accordance with the present invention, the bandwidth of the access link 310 is not equally available to the users 20. Rather the resource arbiter 10 allocates a portion of the bandwidth of the access link 310 to groups of users 20. All users 20 in any single group then have equal access to the allocated portion of bandwidth.

The allocation of any individual user 20 to a group depends on that user's usage intensity of the network, i.e. the average rate of data exchange with the network. Each group of users is allocated a common bandwidth or information transfer rate. The allocation of users to groups is preferably ongoing and variable, such that any user 20 whose usage intensity changes will be reassigned to an appropriate group. Similarly, the distribution of the total bandwidth between the groups is preferably also variable to a certain degree to take account of large changes in the number of users in any one group, or in the average bandwidth usage of any one group. The number of groups should be at least two with the minority of users 20 having a high average information rate being grouped together. If more than two groups are used, those users 20 that only seldom access the network can be grouped together and at least one further group would consist of users having an intermediate level of usage intensity. A mechanism for allocating users to groups and bandwidth proportion to groups is described in further detail below.

Turning now to Fig. 2, the functional elements of the resource arbiter 10 are shown in more detail in accordance with a first embodiment. The resource arbiter 10 includes a multiplexer/demultiplexer 110, to which the user ports 20 are connected. The multiplexer/demultiplexer 110 combines all user traffic into a single physical flow or data stream one one direction and performs the reverse function in the opposite direction. A rate monitor 120 is connected to the multiplexer/demultiplexer 110 and monitors the traffic rate associated with any individual user port 20 and compiles statistics of bandwidth usage. This rate monitoring is performed on the basis of the source or destination of information flowing in both directions through the resource arbiter 10. For example this could be the sending and destination IP addresses of packets. In an Ethernet frame, the physical port and information in the VLAN field may be used to the same effect. The rate monitor is coupled by a data bus carrying the combined data stream to a switch 140. The configuration of the switch 140 is controlled by a switch controller 130, which is coupled to, and receives the rate statistics from, the rate monitor 120. The switch 140 connects data carried on the combined data stream to the available network connections 30 of the access link 310. Based on the traffic statistics for traffic in both directions collected by the rate monitor 120, the switch controller 130 determines to which network connection 30 traffic flows originating at each user port should be connected. This is achieved by routing each outgoing packet, that is each packet originating at one of the users 20, to the allocated network connection 30. It will be appreciated that both the rate monitor 120 and the switch 140 are able to distinguish between packets originating from different user ports 20 in the combined data stream. The multiplexer/demultiplexer 110 also includes buffering (not shown) to temporarily store data for transmission to a network connection and data received over the network and destined for one of the user ports 20 so that user ports 20 will also receive data only at the allocated data transmission rate.

The user ports 20 are divided into groups in accordance with the average information transfer rate as will be described below. Each group of users 20 is allocated a defined portion of the total bandwidth provided by all network connections 30. This proportion may be fixed for a specific network configuration or variable depending on the capacity used by other user groups. In effect the defined proportion of the total bandwidth allocated to any single group is a reserved proportion for that group. Accordingly, this group is guaranteed this proportion of bandwidth when required, however, bandwidth can be "borrowed" from other groups when those groups are not using the allocated bandwidth to full capacity. Within any one group the available bandwidth is allocated to traffic from different users on a fair scheduling basis, such as round robin scheduling or weighted fair queuing.

The multiplexer/demultiplexer 110 allows all processing of the data in the resource arbiter to be performed on a single data stream. While this arrangement reduces the number of physical connections required in the resource arbiter 10, the multiplexer/demultiplexer 110 could be omitted and the rate monitor 120 and switch 140 adapted to handle separate data streams originating from the user ports 20.

The resource arbiter 10 described above with reference to Fig. 2 is capable of restricting traffic originating from any user to a specific shared bandwidth by directing all traffic originating from a group of users 20 to the same network connection 30 or group of network connections 30. The bandwidth division of the access link 310 is thus provided by the physical limitations of the allocated network connections 30. For those cases where only one network connection 30 is available or where the number of network connections 30 does not permit the desired division of bandwidth to ensure that all users obtain the required quality of service, the resource arbiter 10 must effect this division of bandwidth itself by sharing access to one or more network connections 30 between the traffic in each group. This is illustrated in Fig. 3 in accordance with a second embodiment of the present invention. The resource arbiter 10 shown in Fig. 3 includes all elements of that described with reference to Fig. 2. These elements perform the same function as already described so further discussion will be omitted here. In this embodiment only one network connection 30 is shown. As in the first embodiment illustrated in Fig. 2, the switch 140 generates as many connections or group ports 150 from the combined data stream as there are groups of users. The data from each of these group ports must be passed on to the single network connection 30. This is achieved using a resource divider 160, which is connected between the switch 140 and the network connection 30. The resource divider 160 operates only on outgoing traffic, that is, on traffic flowing from a user 20 to the network. The resource divider 160 schedules transmission from each of the group ports so that the transmission capacity of the network connection 30 is divided between these ports. In effect the resource divider time division multiplexes data received through each group port 150 onto the single network connection 30. The resource divider preferably includes buffers for storing packets awaiting scheduled transmission. Depending on the proportion of the network connection bandwidth that is to be allocated to each group port 150, the scheduling scheme may allow each port 150 equal access to the network connection 30, for example using a Round Robin scheduler or alternatively, some other form of time division access control may be used, such as weighted fair queuing.

While the embodiments of the present invention have been described with reference to functional blocks it will be understood that the functions represented by these blocks may be implemented by the use of shared or dedicated hardware, including conventional hardware capable of executing software.

It will be appreciated that the effectiveness of the proposed arrangement depends on the composition of the user groups and also the bandwidth allocated to these groups. A method for performing the initial division of users into two groups is illustrated in Fig. 4. It is assumed at the start of this method that all data transmission to and from all users 20 shares the total bandwidth of the network connections 30. In a first step 200 of Fig. 4, the information rate, or bandwidth usage, of each user port 20 is monitored over a period of time and the average information rate used determined. The longer this period of time, the more effective the division of users into groups will be for all users. However, the initial assignment of users to groups may be determined by a relatively short monitoring period, such as 48 hours or one week. In step 201, the top 1% of users are identified. Preferably, the users are ranked according to average information rate. The top 1% of ranked users are then identified and the average bandwidth or information rate U_T used by this 1% of users is then determined. The figure of 1% is taken by way of example. If the proportion of users having high usage intensity is large, it may be preferable to select a higher percentage of users for the high intensity group. In step 202, the average monitored information rate of all other users, U_R, is determined. In step 203 the demarcation level for the high usage intensity group, i.e. the group containing the users having the highest average information rate, is set. This level is set to the geometric mean of the average information rate U_τ of the 1% most intensive users and the average information rate U_R of all remaining users. In step 204 the information rate of all users is determined after a period of monitoring and a user is allocated to the high usage intensity group if its average information rate is greater than or equal to the demarcation information rate M or to the low usage intensity group if the average information rate is lower. The routing of data can then take place as indicated in step 205. Data from users assigned to the high usage intensity group is routed through a first single or group network connection 30, while data from users assigned to the low usage intensity group is routed through a separate network connection 30 or connections. This routing continues for a predetermined time period T_monitor, after which the usage intensity or average information rate of each user is reassessed and the allocation of users to groups altered if necessary. Although not indicated in Fig. 4, the demarcation level for the high usage intensity group can also be altered after a certain period of time to take account of a change in the usage practices of all users. This is achieved by returning to step 200.

This method for dividing users into groups is only one of several possibilities depending on the purpose of the division. If the purpose is specifically to separate high intensity users from all other users, a preferable method would be to compose a high intensity group of the 1% of highest ranking users as obtained in step 201 of Fig. 4 and all users that have a usage intensity that is close to the average usage intensity U_T of these 1% of the highest ranked users. For example users with a usage intensity that is 90% of the average U_τ could be added to the group, while all other users are in the other group or groups.

In some cases, it may be possible to select the group of highest usage intensity using the average information rate as a limit. In this case, step 201 would be replaced by a step of comparing the information rate of all users with fixed average information rate limits. For example, all users with an average monitored information rate of equal to or more than 50 kbit/s would be placed in the highest usage intensity group. It will be appreciated that a suitable figure is best selected after monitoring the usage patterns of the users over a period of time. The information rate might also be altered with time as the usage practices change.

If three or more groups are required, the method steps 202 and 203 in Fig. 4 are preferably changed and supplemented by further steps. Specifically, prior to determining the average information rate of the remaining users, these users should first be divided into further groups in accordance with the determined usage intensity ranking. The division of users is preferably performed on a percentage basis, such that for example, the 10%> of users having the lowest usage intensity are assigned to a separate group. The remaining users are then assigned to a single group or alternatively a further group divided off. In all cases, the demarcation information rate separating each group should be the mean of the average information rate of the users in each of the bordering groups. A further rule of thumb is that the proportion of users in the highest usage intensity group should be smaller than the proportion of bandwidth allocated to that group. In other words, if 5% of users are assigned to the highest usage intensity group, the proportion of bandwidth allocated to that group must be greater than 5% of the total available bandwidth. It will be appreciated that the absolute value of 1% for determining the most intensive users is a preferable value. Depending on the profile of users in a group a different percentage of users might better characterise the usage intensity for setting a demarcation of the group

An example of how the division of users 20 into groups and allocation of bandwidth to these groups in accordance with the present invention can improve the quality of service is described below.

The access link in this example is an ADSL access link with a total available bandwidth of 20 Mbit/s and having 2000 users, each user having an individual bandwidth limit of 2 Mbit/s. It is assumed in this example that the bandwidth limit for the download of large files is within the capacity of the access link and that during an Internet surfing session an average bit rate of 100 kbit/s is used in a very bursty manner, i.e. 5% of the time at 2Mbit/s, if permitted by the network.

Taking first the case when all users share the combined bandwidth of the access link, if at any one time 100 users are surfing the Internet while 100 other users are downloading large files, each user will at best, that is under a normal best effort scheduling scheme, obtain a data rate of 190kbit/s. While this is a reasonable speed for file download, particularly for a low cost ADSL line, this data rate is unfair to Internet surfers. The Internet surfers constitute 50% of the active users yet obtain less than 2.5% of the bits transferred.

Moreover the system response time will be slower giving a low quality of service and leading to disgruntled customers.

By applying the user grouping scheme in accordance with the present invention, the situation of those users who request only an intermittent high information rate from the network will be considerably improved. By creating two groups of users and reserving an information rate of 12Mbit/s for the 100 intensive network users and the remaining 8Mbit/s for the 1900 less intensive users any one of the intensive users downloading a file will obtain a data rate of 120kbit/s, while any one of the 100 Internet surfers in the other group will obtain an information rate of 80 kbit/s on average. Assuming as before that information is transferred only 5% of the time, this gives a bandwidth of 1.6Mbit/s. This system thus provides a much improved service to the users who predominately surf the Internet. It is further important to note that the largest proportion of bandwidth has been assigned to the group containing the most intensive users. These users thus benefit from a reasonable information rate for downloading files. It is true that any user in this group attempting to obtain higher data transfer rates intermittently, such as is required for surfing the Internet, will experience a greater or lesser degree of latency depending on the usage of other users in the group, however, any disadvantage experienced in this type of usage is largely offset by the value extracted from the service by the large amounts of average data transfer. This grouping is thus substantially fairer than when intensive and much less intensive users share the same bandwidth. Moreover, those customers who are effectively paying more for each bit of data transfer and are thus of most value to ISPs are provided with an acceptable quality of service. The improved quality of service is maintained for all users with the exception of the minority of very intensive users when more bandwidth is allotted to groups with more intensive users. Moreover, even though the more intensive users no longer have access to the whole shared bandwidth of the access link they will nevertheless have access to a larger common bandwidth than users in other groups, so cannot be seen to have been penalised by this scheme.

It is important when setting up the user groups to ensure that the most intensive users are isolated from other users as much as possible. Any user that is grouped with intensive users will need to use the access link intensively himself to obtain value from the service. It is clear that a user that obtains only intermittent high information transfer will be at a disadvantage in a group containing users that demand a more continuous information transfer rate. It is possible to allow the groups of more intensive users to "borrow" bandwidth from groups containing less intensive users when this bandwidth is unused.

When the system is set up, it is likely that some users will initially be placed in the wrong group either to their benefit or disadvantage. However, since the usage intensity of the users 20 is preferably monitored continuously by the rate monitor 120 a true picture of a users average information transfer rate can be formed over time. The sorting mechanism can thus continually adapt to the changing habits of the users by reassigning users to groups when a user is determined to have a significantly different usage pattern from the average usage intensity of the group. Preferably this reassessment of the users allocation to groups is performed at intervals. By making these intervals variable, for example in a stochastic manner, users will be prevented from falsifying the usage intensity statistics by modifying their use of the network prior to a reassessment. Naturally, this could also be avoided by performing the monitoring continuously and determining an average usage intensity over the time since the last evaluation.

In addition to selecting an appropriate bandwidth usage demarcation level for determining which users are allocated to which group, it is further advantageous to build in a hysteresis into the allocation mechanism to prevent a too frequent reassignment of a user to different groups.

In addition to monitoring the average information rate obtained by each user, the rate monitor may weight the information rate. For example, information sent or received during peak periods, such as office hours would preferably be weighted more heavily than usage occurring during the night. In this manner a bit received during business hours could be counted as equal to four bits received at other times. In addition, weighting may also be applied to the usage according to the distance travelled, for example with information that has travelled over a transatlantic link being weighted more heavily than information that has used only domestic or continental links. This would also assist in making the service available to all users more equitable by evening out the value extracted from the service.

Claims

Claims:

1. In a packet-switched network access arrangement, a method for allocating transmission bandwidth to network users, including: monitoring the rate of information exchange to and from each user, characterised by the steps of providing at least two channels in said access arrangement, wherein each channel has a defined reserved proportion of the total transmission bandwidth, dividing the users into at least two groups as a function of the monitored information exchange rate such that users with a similar information exchange rate are in the same group; assigning a channel to each group of users such that information exchange for all users of a group are routed through the assigned channel and each user in a group has access to the allocated proportion of bandwidth.

2. A method as claimed in claim 1, characterised by the step of: allocating a larger proportion of the total bandwidth to channels used by user groups having the highest rate of information exchange.

3. A method as claimed in claim 1 or 2, characterised by the further step of: performing said monitoring step over a period of time and determining an average information exchange rate over said period of time.

4. A method as claimed in any previous claim, characterised by at intervals reforming the groups of users if the monitored information exchange of at least some users has changed.

5. A method as claimed in claim 4, characterised in that said intervals are varied stochastically.

6. A method as claimed in any previous claim, characterised in that said plurality of channels are time division multiplexed onto a single link.

7. A method as claimed in any previous claim, characterised by routing information exchange of one group through a second channel assigned to a further group of users, when bandwidth is available in said second channel.

8. A method as claimed in any previous claim, characterised by permitting each user in a group equal access to the allocated proportion of bandwidth.

9. A method as claimed in claim 8, characterised by distributing access for each user in a group to the allocated proportion of bandwidth using round- robin scheduling.

10. A method as claimed in any previous claim, characterised by providing the users in a group unequal access to the allocated proportion of bandwidth.

11. A method as claimed in claim 10, characterised by distributing access for each user in a group to the allocated proportion of bandwidth using weighted fair queuing.

12. An arrangement for allocating bandwidth to users of a packet-switched network said arrangement including a rate monitor (120) for measuring the rate of information transfer to and from a plurality of user ports (20), characterised by an access link (310) including a plurality of defined bandwidth channels (30; 150) for connecting said user ports to said network, a switch assembly (140) for routing data between each user port and said network through one of said defined-bandwidth channels, a switch controller (130) coupled to said rate monitor and said switch assembly and arranged to control the routing of data through said switch assembly as a function of the monitored rate of information transfer of each user, such that data for users having a high rate of information transfer is routed through one defined-bandwidth channel and data for users having a low rate of information transfer is routed through a separate defined-bandwidth channel.

13. An arrangement as claimed in claim 12, further characterised by a multiplexer (110) coupled to said rate monitor for combining data to and from said user ports into a single data stream.

14. An arrangement as claimed in claim 12 or 13, characterised by a time division multiplexer (160) coupled to said switch for time division multiplexing user data from at least two channels onto a single network connection.