1. 1 Netcomm 2005 Communication Networks Recitation 5

2. 2 Netcomm 2005 Input Queuing Scheduling & Combined Switches

3. 3 Netcomm 2005 Output-queued switches Best delay and throughput performance - Possible to erect “bandwidth firewalls” between sessions Main problem - Requires high fabric speedup (S = N) Unsuitable for high-speed switching

4. 4 Netcomm 2005 Input-queued switches Big advantage - Speedup of one is sufficient Main problem - Can’t guarantee delay due to input contention Overcoming input contention: use higher speedup

5. 5 Netcomm 2005 Input Queue Scheduling First Goal : maximize throughputFirst Goal : maximize throughput Second Goal : control packet delaySecond Goal : control packet delay Methods :Methods : –maximum matching –maximal matching –maximum/maximal weight matching –stable matching using Virtual Output Queuing and moderate speedup

6. 6 Netcomm 2005 Example of Maximal Size Matching A1 B C D E F 2 3 4 5 6 A1 B C D E F 2 3 4 5 6 Maximal Size Matching Maximum Size Matching A B C D E F 1 2 3 4 5 6

7. 7 Netcomm 2005 Parallel Iterative Matching 1 2 3 4 1 2 3 4 Requests 1 2 3 4 1 2 3 4 Grant 1 2 3 4 1 2 3 4 Accept/Match 1 2 3 4 1 2 3 4 #1 #2 Random Selection 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4

8. 8 Netcomm 2005 Parallel Iterative Matching Convergence Time Number of iterations to converge:

9. 9 Netcomm 2005 iSLIP 1 2 3 4 1 2 3 4  1: Requests 1 2 3 4 1 2 3 4  2: Grant 1 2 3 4 1 2 3 4  3: Accept/Match 1 2 3 4 1 2 3 4 #1 #2 Round-Robin Selection 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4

10. 10 Netcomm 2005 iSLIP Properties Random under low loadRandom under low load TDM under high loadTDM under high load Lowest priority to Most Recently UsedLowest priority to Most Recently Used 1 iteration: fair to outputs1 iteration: fair to outputs Converges in at most N iterations.Converges in at most N iterations. On average < log 2 N.On average < log 2 N.

11. 11 Netcomm 2005 Input Queueing Longest Queue First or Oldest Cell First Weight Waiting Time 100% Queue Length { } =

12. 12 Netcomm 2005 Input Queueing Why is serving long/old queues better than serving maximum number of queues? When traffic is uniformly distributed, servicing the maximum number of queues leads to 100% throughput. When traffic is non-uniform, some queues become longer than others. A good algorithm keeps the queue lengths matched, and services a large number of queues. VOQ # Avg Occupancy Uniform traffic VOQ # Avg Occupancy Non-uniform traffic

13. 13 Netcomm 2005 Speedup: Context MemoryMemory MemoryMemory The placement of memory gives - Output-queued switches - Input-queued switches - Combined input and output queued switches A generic switch

14. 14 Netcomm 2005 Using Speedup 1 1 1 2 2

15. 15 Netcomm 2005 The Speedup Problem Find a compromise: 1 < Speedup << N - to get the performance of an OQ switch - close to the cost of an IQ switch Essential for high speed QoS switching

16. 16 Netcomm 2005 What is exact mimicking? Apply same inputs to an OQ and a CIOQ switch - packet by packet Obtain same outputs - packet by packet Key concept: urgency value - urgency = departure time - present time

17. 17 Netcomm 2005 Most Urgent Cell First (MUCF) The algorithm - Outputs try to get their most urgent packets - Inputs grant to output whose packet is most urgent, ties broken by port number - Loser outputs for next most urgent packet - Algorithm terminates when no more matches are possible Speedup of 4 is sufficient for exact emulation of FIFO OQ switches, with MUCF (Prabhakar & McKeown, 1997) Joined Preferred Matching (JPM) – Speedup 2 (Stoica & Zhang, 1998)