Method to achieve bounded buffer sizes and quality of service guarantees in the internet network
Methods to achieve bounded router buffer sizes and Quality of Service guarantees for traffic flows in a packet-switched network are described. The network can be an Internet Protocol (IP) network, a Differentiated Services network, an MPLS network, wireless mesh network or an optical network. The routers can use input queueing, possibly in combination with crosspoint queueing and/or output queueing. Routers may schedule QoS-enabled traffic flows to ensure a bounded normalized service lead/lag. Each QoS-enabled traffic flow will buffer O(K) packets per router, where K is an integer bound on the normalized service lead/lag. Three flow-scheduling methods are analysed. Non-work-conserving flow-scheduling methods can guarantee a bound on the normalized service lead/lag, while work-conserving flow-scheduling methods typically cannot guarantee the same small bound. The amount of buffering required in a router can be reduced significantly, the network links can operate near peak capacity, and strict QoS guarantees can be achieved.
1. A switch for transmitting a plurality of traffic flows over a scheduling frame with duration F seconds for F>0 with a plurality of transmission intervals, wherein each of the traffic flows belongs to either a first traffic class or a second traffic class, and wherein each of the traffic flows has an associated data rate, comprising:
N input ports, M output ports for integers N>0 and M>0;
N×M Virtual Queues (VQs) for buffering data associated with the traffic flows, wherein each of the N×M VQs is associated with an input port and an output port, and M of the VQs are associated with each one of the N input ports, wherein each of the traffic flows is associated with one VQ, and wherein each of said VQs is associated with a guaranteed data rate;
wherein each of said VQs is further partitioned into a first sub-queue and a second sub-queue, wherein said first sub-queue buffers data associated with those of the traffic flows belonging to the first traffic class, wherein said second sub-queue buffers data associated with those of the traffic flows belonging to the second traffic class, and wherein each of said sub-queues is associated with a guaranteed data rate;
memory for storing a first schedule, wherein said first schedule specifies for each input port which associated VQ, if any, has a reservation to transmit data to its associated output port in each transmission interval in said scheduling frame;
memory for storing a second schedule, wherein said second schedule specifies for each of said VQs which has a reservation to transmit data in a transmission interval in said scheduling frame, which of its associated sub-queues, if any, has a reservation to transmit data in said transmission interval;
wherein the first schedule provides each of said VQs with a guaranteed rate of transmission through said switch in each scheduling frame, sufficient to satisfy its guaranteed data rate requirement;
wherein the second schedule provides each of the first and second sub-queues associated with each of said VQs with a guaranteed rate of transmission through said switch in each scheduling frame, sufficient to satisfy its guaranteed data rate requirement; and
wherein for each VQ with R reservations per scheduling frame, said first schedule provides that VQ with at least R/2−K reservations and at most R/2+K reservations in each half of said scheduling frame with duration F/2 seconds, wherein integer K is the bound on the service lead and service lag measured in reservations per scheduling frame, and wherein R is substantially larger than K.
2. A switch for switching a plurality of guaranteed-rate (GR) traffic flows over a set of output ports, given a scheduling frame comprising a plurality of transmission intervals defined over a duration of time of F seconds for real number F, said switch comprising:
N input ports for receiving data and M output ports for transmitting data, for integers N and M;
a plurality of queues, wherein each queue buffers data which arrives at a common one of said N input ports and which departs on a common one of said M output ports; wherein each GR traffic flow is associated with a guaranteed data-rate requirement, wherein each GR traffic flow is associated with one queue, and wherein data associated with a GR traffic flow is buffered in its associated queue;
memory for storing a flow-schedule,
wherein the flow-schedule determines which of the GR traffic flows associated with said queues, if any, have a reservation to transmit in each transmission interval in said scheduling frame;
wherein the flow-schedule provides each of said GR traffic flows with a guaranteed number of reservations per scheduling frame sufficient to meet its guaranteed data-rate requirement,
and wherein for each GR traffic flow with R reservations per scheduling frame for integer R, said flow-schedule provides said GR traffic flow with a least R/2−K reservations and at most R/2+K reservations in each half of said scheduling frame with a duration of F/2 seconds, where integer K is the bound on the service lead and service lag measured in reservations per scheduling frame, and wherein R is substantially larger than K.
3. The switch of claim 2 ,
wherein for each GR traffic flow with R reservations per scheduling frame, said flow-schedule provides said GR traffic flow with a least R/4−K reservations and at most R/4+K reservations in each quarter of said scheduling frame with a duration of F/4 seconds.
4. The switch of claim 3 ,
wherein for each GR traffic flow with R reservations per scheduling frame, said flow-schedule provides said GR traffic flow with at least R*(E/F)−K reservations and at most R*(E/F)+K reservations in the subset of the scheduling frame spanning time 0 to E, wherein E<F, and wherein E is a real number.
5. The switch of claim 3 , wherein the memory is distributed throughout the switch.
6. The switch of claim 3 , wherein the GR traffic flows belong to a new traffic class representing Guaranteed-Rate traffic, equivalently Deterministic, traffic, in a Differentiated Services service model.
7. The switch of claim 3 , further comprising:
a traffic-shaper module, said module comprising:
a controller, memory to buffer data, memory to buffer tokens, and a token-generator;
wherein said traffic-shaper module is operable to buffer incoming data associated with at least one of the GR traffic flows, and release said data in a manner to reduce the burstiness of said GR traffic flow.
8. The switch of claim 3 , wherein each of said queues is partitioned into a plurality of flow-queues, and wherein each flow-queue buffers data associated with one GR traffic flow.
9. The switch of claim 4 , wherein each of said queues is partitioned into a plurality of flow-queues, and wherein each flow-queue buffers data associated with one GR traffic flow.
10. The switch of claim 3 , further comprising:
memory for storing a queue-schedule, wherein each of said plurality of queues is associated with a guaranteed data-rate requirement;
wherein said queue-schedule determines which of said queues, if any, have a reservation to transmit in each transmission interval of said scheduling frame,
and wherein said queue-schedule provides each of said queues with a guaranteed number of reservations per scheduling frame sufficient to satisfy its guaranteed data-rate requirement.
11. The switch of claim 10 , wherein for each queue with R reservations per scheduling frame, said queue-schedule provides that queue with at least R/2−K reservations and at most R/2+K reservations in each half of said scheduling frame with duration F/2 seconds.
12. The switch of claim 11 ,
wherein for each queue with R reservations per scheduling frame, said queue-schedule provides that queue with at least R/4−K reservations and at most R/4+K reservations in each quarter of said scheduling frame with duration F/4 seconds.
13. A switch for switching a plurality of class-based (CB) traffic flows over a set output ports, given a scheduling frame comprising a plurality of transmission intervals defined over a duration of time of F seconds for real number F, said switch comprising:
N input ports for receiving data and M output ports for transmitting data, for integers N and M;
a plurality of class-queues, wherein each class-queue buffers data associated with one class of traffic;
wherein each class-queue buffers data which arrives at a common one of said N input ports and departs on a common one of said M output ports;
wherein each class-queue is associated with a guaranteed data-rate requirement;
wherein each CB traffic flow is associated with one class of traffic and with one class-queue;
wherein data associated with a CB traffic flow is buffered in its associated class-queue; memory far storing a class-schedule;
wherein the class-schedule determines which of said class-queues, if any, has a reservation to transmit in each transmission interval in said scheduling frame; and
wherein said class-schedule provides each of said class-queues with a guaranteed number of reservations per scheduling frame sufficient to meet its guaranteed data-rate requirement; and
wherein for each class-queue with R reservations per scheduling frame for integer R>=2, said class-schedule provides that class-queue at least R/2−K reservations and at most R/2+K reservations in each half of said scheduling frame with duration F/2 seconds, wherein integer K is the bound on the service lead and service lag measured in reservations per scheduling frame, and wherein R is substantially larger than K.
14. The switch of claim 13 , further comprising: memory to store a class-schedule; wherein each class-queue is associated with a guaranteed data-rate requirement, wherein said class-schedule determines which of said class-queues, if any, has a reservation to transmit in each transmission interval in said scheduling frame; and wherein said class-schedule provides each of said class-queues with a guaranteed number of reservations per scheduling frame sufficient to meet its guaranteed data-fate requirement.
15. The switch of claim 14 ,
wherein for each class-queue with R reservations per scheduling frame, said class-schedule provides that class-queue with at least R/2−K reservations and at most R/2+K reservations in each half of said scheduling frame with duration F/2 seconds.
16. The switch of claim 15 ,
wherein for each class-queue with R reservations per scheduling frame, said class-schedule provides that class-queue with at least R/4−K reservations and at most R/4+K reservations in each quarter of said scheduling frame with duration F/4 seconds.
17. The switch of claim 16 ,
wherein for each class-queue with R reservations per scheduling frame, said class-schedule provides said class-queue with at least R*(E/F)−K reservations and at most R*(E/F)+K reservations in the subset of the scheduling frame spanning time 0 to E, wherein E is real number and E<F.
18. The switch of claim 14 , wherein at least one of the CB traffic flows belongs to a traffic class in a Differentiated Services service model, said classes including an Expedited Forwarding (EF) traffic class, an Assured Forwarding (AF) traffic class, and a Best Effort (BE) traffic class.
19. The switch of claim 15 , further comprising a set of class-queues for at least one traffic, class in a Differentiated Services (DiffServ) service model, said classes including an Expedited Forwarding (EF) traffic class, an Assured Forwarding (AF) traffic class, and a Best Effort (BE) traffic class.
20. The switch of claim 15 , wherein the GE traffic flows belong to a new traffic class representing Guaranteed-Rate traffic, equivalently Deterministic traffic, in a Differentiated Services (DiffServ) service model.
21. The switch of claim 15 , further comprising a scheduling processor, wherein said scheduling-processor will process the contents of a class-queue, to determine which data buffered within said class-queue is selected to transmit next.
22. The switch of claim 21 , wherein said scheduling-processor selects the oldest data in said class-queue to transmit next.
23. The switch of claim 21 ,
wherein said scheduling-processor selects the data with the smallest ‘Virtual Finishing Time’ (VFT) in said class-queue to transmit next, wherein the VFT is assigned to data arriving to said class-queue using the Generalized Processor Sharing/Weighted Fair Queuing (GPS/WFQ) algorithms developed by Parekh and Gallager.
24. The switch of claim 3 , wherein all transmission intervals within the scheduling frame have the same duration.
25. The switch of claim 3 , wherein at least 2 transmission intervals within the scheduling frame have different durations.
26. The switch of claim 3 , wherein R>=64 and K does not exceed the square-root of R.
27. The switch of claim 3 , wherein R>=16 and K=4.
28. The switch of claim 3 , wherein the data to be transmitted is formatted as fixed-sized cells.
29. The switch of claim 3 , wherein the data to be transmitted is formatted as Internet Protocol (IP) packets with variable sizes.
30. The switch of claim 3 , wherein the data to be transmitted is formatted as MPLS packets.
31. The switch of claim 3 , wherein the data to be transmitted is formatted as Ethernet packets.