Presentation is loading. Please wait.

Presentation is loading. Please wait.

D2.2 Report on Movement in Switch State of the Artand Commercial Motivators (RECN) Annex A (June 7th 2006) Ian Johnson Xyratex.

Similar presentations


Presentation on theme: "D2.2 Report on Movement in Switch State of the Artand Commercial Motivators (RECN) Annex A (June 7th 2006) Ian Johnson Xyratex."— Presentation transcript:

1 D2.2 Report on Movement in Switch State of the Artand Commercial Motivators (RECN) Annex A (June 7th 2006) Ian Johnson Xyratex

2 A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistage Interconnection Networks J. Duato 1, I. Johnson 2, J. Flich 1, F. Naven 2, P.J. García 3, T. Nachiondo 1 1 Technical University of Valencia Valencia, Spain 2 Xyratex Havant, UK 3 University of Castilla-La Mancha Albacete, Spain The Eleventh International Symposium on High-Performance Computer Architecture, San Francisco, 2005

3 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 3 Outline Introduction Congestion and HOL blocking Why now? Why previous proposals are inadequate Proposal: RECN Performance evaluation Conclusions

4 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 4 Interconnection Networks MPPs Earth Simulator (640 vectorial CPUs) ASCI Q (12,288 EV68 CPUs, Quadrics network) BlueGene/L (65.535 nodes, each one 2 processors, 360 TFlops) PC Clusters Storage Area Network (SANs) –Google (6.000 CPUs and 12.000 disks) Thunder (1.024 nodes each one 4 Itaniums/8GB) Many data centers all around the world ASCI Q Earth Simulator Thunder

5 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 5 Network Throughput beyond Saturation

6 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 6 Network Contention Several packets request the same output port One makes progress, the others wait Network Congestion Persistent network contention It is quickly propagated by flow control (lossless nets) Network performance degrades dramatically Head of line (HOL) blocking When the first packet in a queue is blocked, any other packet in the same queue is also blocked, even if it will request available resources Congestion and HOL Blocking

7 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 7 Congestion and HOL Blocking Network contention

8 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 8 Congestion and HOL Blocking Persistent network contention

9 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 9 Congestion and HOL Blocking Persistent network contention Flow control

10 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 10 Congestion and HOL Blocking Persistent network contention Congestion propagates

11 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 11 Congestion and HOL Blocking Congestion introduces HOL blocking, and this may degrade network performance dramatically 33% HOL 33% 33% 100% 33% 100%

12 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 12 Traditional Solution Overdimensioning the network Latency Injected traffic Congestion zone Working zone Network bandwidth is much higher than the bandwidth requested by end nodes Low link utilization

13 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 13 Why Congestion Management Now? New problems arising: System cost: Recent interconnects (Myrinet, InfiniBand, ASI) are expensive compared to processors Power consumption: As network size increases, higher power consumption, higher heat dissipation Frequency/voltage scaling techniques: Not very efficient, and do not solve the system cost problem Possible Solutions: Reducing the number of network components: Possible by using a suitable topology, but link utilization increases Systems will work closer to network saturation zone, thus, a congestion management technique will be mandatory

14 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 14 Why Current Techniques Are not Suitable? Proactive Congestion Management (congestion prevention) Path setup before data transmission Used in ATM, computer networks (QoS) High overhead, high latencies (not suitable for HPC) The real problem is not the congestion, but its negative effects (HOL blocking) Reactive Congestion Management (congestion recovery) Injection limitation techniques using closed-loop feedback Do not scale with network size and link bandwidth –Notification delay (proportional to distance) –Link capacity (proportional to clock frequency) –May produce network instabilities

15 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 15 Why Current Techniques Are not Suitable? HOL blocking elimination/reduction DAMQs and Virtual Channels –not efficient for multihop networks VOQ (Virtual Output Queueing) –VOQ at switch level scales but does not eliminate HOL –VOQ at network level: A separate queue at every input port for every destination –Number of required resources scales at least quadratically with network size !!! Credit Flow Controlled ATM –References congestion to network output only –Consumes large number of buffers: A separate queue at every output port for every destination

16 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 16 Proposal Initial idea: Exploit spatial and temporal locality in packet destinations Manage the set of queues as a cache –No equivalent to main memory!!! (where to replace?) –Not enough locality!!! (reduction in queue silicon area by a factor of 4) Observation: Non-congested flows do not introduce significant HOL blocking RECN: Regional Explicit Congestion Notification Non-congested flows are mapped to the same queue Effective reduction in number of queues and no replacement needed Congested flows are detected and mapped to set aside queues (SAQs) RECN is a scalable congestion management technique because: It reacts locally (and thus, it is not affected by propagation delays) A very small number of queues (SAQs) for a wide range of network sizes RECN enables: Effective reduction of network cost by working closer to the saturation point More efficient use of voltage/frequency scaling techniques

17 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 17 RECN Based on the PCI Express Advanced Switching Interconnect (ASI) specification Routing (turnpools) Relevant switch architectural features Congestion detection Congestion notification and queue allocation Queue deallocation Packet processing Flow control

18 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 18 Turnpools 0 1 2 3 4 5 6 7 3 7 turn pool t. pointerD Direction bit Turn example 31 bits = 2 31 destinations AS packet header 2 1 1 2 2 1 1 3 A A 3 3 2 2 B B 2 Allows to know if a packet will pass through a given port in the network Mask bits required

19 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 19 Switch Model RAM in XBAR S=1.5 RAM in RAM in... RAM out RAM out RAM out Arbiter... Dynamic queue management (VCs) Dynamic queue management (VCs) LC

20 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 20 RAM in and RAM out RAM SAQ 0 SAQ 1 SAQ 2 SAQ 3 Cold Queue SAQ 0 SAQ 1 SAQ 2 SAQ 3 Tokens (one per each input port) Root Only at egress: Avoids successive internal notifications v v v v v v v v turn pool mask bits CAM SAQ 0 SAQ 1 SAQ 2 SAQ 3 b b b b b b b b Valid bit Congested point blocked nextSAQ Xon/Xoff Flow control Xoff lv leave bit (only at ingress)

21 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 21 A congestion point forms How it Works

22 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 22 How it Works Cold queue fills over a threshold

23 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 23 How it Works Congestion Detection: Cold Queue at output port side fills over Detection Threshold Congested point: output port SAQs are not allocated at the output port

24 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 24 How it Works

25 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 25 How it Works Internal notification to each input port sending packets to the output port

26 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 26 How it Works Congestion Information Notification: Congestion is notified to input ports sending packets to congested ports Notification includes turnpool information and mask bits Root token set for the input port

27 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 27 How it Works

28 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 28 How it Works Input ports allocate a new SAQ for packets addressed to the congested output port

29 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 29 How it Works Actions after receiving notification: A new SAQ is allocated The notified Turnpool and Mask bits are used to map the new SAQ

30 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 30 How it Works Reception of packets after mapping SAQs (Example 1): 0 0 0 0 1 1 1 1...0003500......0003000.....00011110....000110.. SAQ 0 SAQ 1 SAQ 2 SAQ 3 SAQ 0 SAQ 1 SAQ 2 SAQ 3 Cold Queue 3 36 4 5 3 3 36...0003500......0003000.....00011110....000110.. SAQ 0 ? ? 3 *

31 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 31 How it Works Reception of packets after mapping SAQs (Example 2): 0 0 0 0 1 1 1 1...0003500......0003000.....00011110....000110.. SAQ 0 SAQ 1 SAQ 2 SAQ 3 SAQ 0 SAQ 1 SAQ 2 SAQ 3 Cold Queue * 4 92 4 5 4 4 92...0003500......0003000.....00011110....000110.. COLD ? ?

32 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 32 How it Works

33 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 33 How it Works Notification sent when the SAQ fills over a threshold

34 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 34 How it Works RECN 0 0 0 0 0 0 1 1...0003500.....00011110.. SAQ 0 SAQ 1 SAQ 2 SAQ 3 SAQ 0 SAQ 1 SAQ 2 SAQ 3 Cold Queue...000500.....00011110.. 5 S0 0 0 0 0 0 0 0 0 No leaf Congestion propagation: A RECN packet including turn pool, mask bits, and SAQ id is sent

35 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 35 How it Works

36 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 36 How it Works A new SAQ allocated for the congested port at each output port

37 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 37 How it Works Internal notification when the SAQ fills over A threshold

38 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 38 How it Works The input port allocates A new SAQ

39 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 39 How it Works At the end, the congestion tree builds and is mapped entirely onto SAQs

40 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 40 Performance Evaluation Evaluation based on simulation results Two evaluation studies: Network performance when using: –RECN –VOQ at network level (VOQnet) –VOQ at switch level (VOQsw) –4 queues at ingress and egress ports (4Q) –1 queue at ingress and egress ports (1Q) RECN scalability

41 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 41 Simulation Model Network configurations evaluated: 64 hosts connected by a 64x64 BMIN 256 hosts connected by a 256x256 BMIN 512 hosts connected by a 512x512 BMIN Simulation assumptions: BMINs based on shuffle-exchange connection scheme Deterministic routing 128 KB memories at ingress/egress ports Multiplexed crossbar (BW=12 Gbps) Serial full-duplex pipelined links (BW=8 Gbps) 64 and 512-byte packets Credit-based and Xon-Xoff (for SAQs) flow control Maximum of 8 SAQs at ingress/egress ports (RECN)

42 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 42 Traffic Load Synthetic Traffic: Traces: From I/O activity at cello system disk interface Different compression factors applied # SrcsDst. Injection Rate (%) Traffic Start Time Traffic End Time Corner Case 1 75%Random50%0Sim. End 25%Hot-Spot100%800 μs970 μs Corner Case 2 75%Random100%0Sim. End 25%Hot-Spot100%800 μs970 μs

43 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 43 Performance Comparison Network throughput - Corner case 1, 64x64 BMIN

44 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 44 Performance Comparison Network throughput - Corner case 2, 64x64 BMIN

45 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 45 Performance Comparison Network throughput – Traces, 64x64 BMIN Compression Factor set to 20Compression Factor set to 40

46 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 46 Scalability Analysis SAQ utilization – Corner Case 1, 64x64 BMIN Maximum # SAQs used (ingress) Maximum # SAQs used (egress) Total # of active SAQS

47 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 47 Scalability Analysis SAQ utilization – Corner Case 2, 64x64 BMIN Maximum # SAQs used (ingress) Maximum # SAQs used (egress) Total # of active SAQS

48 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 48 Scalability Analysis SAQ utilization – Traces, Comp. Factor 20, 64x64 BMIN Maximum # SAQs used (ingress) Maximum # SAQs used (egress) Total # of active SAQS

49 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 49 Scalability Analysis SAQ utilization – Traces, Comp. Factor 40, 64x64 BMIN Maximum # SAQs used (ingress) Maximum # SAQs used (egress) Total # of active SAQS

50 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 50 Scalability Analysis Network throughput – Corner Case 2, 256x256 BMIN Maximum # SAQs used (egress) Maximum # SAQs used (ingress)

51 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 51 Scalability Analysis Network throughput – Corner Case 2, 512x512 BMIN Maximum # SAQs used (ingress) Maximum # SAQs used (egress)

52 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 52 Final Remarks We also designed a protocol to deallocate SAQs when they are no longer needed Many optimizations –CAM IDs to reduce control message size –CAM search done in parallel with packet reception –Merging of congestion trees Silicon area reduced with respect to switch- level VOQs

53 Title:A New Scalable and Cost-Effective Congestion Management Strategy for Lossless Multistate Int. Networks Conference: The 11th International Symposium on Hogh-Performance Computer Architecture 53 Conclusions We have proposed a scalable congestion management strategy for lossless networks We have shown that it only requires a small number of buffers for a wide range of network sizes We have modeled an existing ASI switch design, verifying: –Maintains network performance close to ideal (but non-scalable) solution –Silicon area requirements are now smaller than for the original design


Download ppt "D2.2 Report on Movement in Switch State of the Artand Commercial Motivators (RECN) Annex A (June 7th 2006) Ian Johnson Xyratex."

Similar presentations


Ads by Google