On the Correct Sizing on Meshes Through an Effective Congestion Management Strategy P. J. García 1, J. Flich 2, J. Duato 2, I. Johnson 3, F. J. Quiles.

Slides:



Advertisements
Similar presentations
1 ZonicBook/618EZ-Analyst Resonance Testing & Data Recording.
Advertisements

Variations of the Turing Machine
1 EE384Y: Packet Switch Architectures Part II Load-balanced Switch (Borrowed from Isaac Keslassys Defense Talk) Nick McKeown Professor of Electrical Engineering.
EE384y: Packet Switch Architectures
1 UNIT I (Contd..) High-Speed LANs. 2 Introduction Fast Ethernet and Gigabit Ethernet Fast Ethernet and Gigabit Ethernet Fibre Channel Fibre Channel High-speed.
AP STUDY SESSION 2.
1
1 Vorlesung Informatik 2 Algorithmen und Datenstrukturen (Parallel Algorithms) Robin Pomplun.
Copyright © 2003 Pearson Education, Inc. Slide 1 Computer Systems Organization & Architecture Chapters 8-12 John D. Carpinelli.
Part 3 Probabilistic Decision Models
McGraw-Hill©The McGraw-Hill Companies, Inc., 2003 Chapter 11 Ethernet Evolution: Fast and Gigabit Ethernet.
Copyright © 2011, Elsevier Inc. All rights reserved. Chapter 6 Author: Julia Richards and R. Scott Hawley.
Author: Julia Richards and R. Scott Hawley
1 Copyright © 2013 Elsevier Inc. All rights reserved. Chapter 3 CPUs.
Properties Use, share, or modify this drill on mathematic properties. There is too much material for a single class, so you’ll have to select for your.
UNITED NATIONS Shipment Details Report – January 2006.
RXQ Customer Enrollment Using a Registration Agent (RA) Process Flow Diagram (Move-In) Customer Supplier Customer authorizes Enrollment ( )
1 Hyades Command Routing Message flow and data translation.
Business Transaction Management Software for Application Coordination 1 Business Processes and Coordination. Introduction to the Business.
1 RA I Sub-Regional Training Seminar on CLIMAT&CLIMAT TEMP Reporting Casablanca, Morocco, 20 – 22 December 2005 Status of observing programmes in RA I.
Properties of Real Numbers CommutativeAssociativeDistributive Identity + × Inverse + ×
Create an Application Title 1A - Adult Chapter 3.
FACTORING ax2 + bx + c Think “unfoil” Work down, Show all steps.
Rhesy S.ppt proRheo GmbH
Evaluating Window Joins over Unbounded Streams Author: Jaewoo Kang, Jeffrey F. Naughton, Stratis D. Viglas University of Wisconsin-Madison CS Dept. Presenter:
1 Click here to End Presentation Software: Installation and Updates Internet Download CD release NACIS Updates.
1 Maintaining Packet Order in Two-Stage Switches Isaac Keslassy, Nick McKeown Stanford University.
REVIEW: Arthropod ID. 1. Name the subphylum. 2. Name the subphylum. 3. Name the order.
1 EE 122: Networks Performance & Modeling Ion Stoica TAs: Junda Liu, DK Moon, David Zats (Materials with thanks.
Break Time Remaining 10:00.
Mohamed Hauter CMPE 259 – Sensor Networks UCSC 1.
D2.2 Report on Movement in Switch State of the Artand Commercial Motivators (RECN) Annex A (June 7th 2006) Ian Johnson Xyratex.
Turing Machines.
Table 12.1: Cash Flows to a Cash and Carry Trading Strategy.
Local Area Networks - Internetworking
Advance Nano Device Lab. Fundamentals of Modern VLSI Devices 2 nd Edition Yuan Taur and Tak H.Ning 0 Ch9. Memory Devices.
PP Test Review Sections 6-1 to 6-6
1 Atomic Routing Games on Maximum Congestion Costas Busch Department of Computer Science Louisiana State University Collaborators: Rajgopal Kannan, LSU.
EU market situation for eggs and poultry Management Committee 20 October 2011.
Hash Tables.
EIS Bridge Tool and Staging Tables September 1, 2009 Instructor: Way Poteat Slide: 1.
Mohamed ABDELFATTAH Vaughn BETZ. 2 Why NoCs on FPGAs? Embedded NoCs Power Analysis
Outline Minimum Spanning Tree Maximal Flow Algorithm LP formulation 1.
Exarte Bezoek aan de Mediacampus Bachelor in de grafische en digitale media April 2014.
Copyright © 2012, Elsevier Inc. All rights Reserved. 1 Chapter 7 Modeling Structure with Blocks.
1 RA III - Regional Training Seminar on CLIMAT&CLIMAT TEMP Reporting Buenos Aires, Argentina, 25 – 27 October 2006 Status of observing programmes in RA.
Basel-ICU-Journal Challenge18/20/ Basel-ICU-Journal Challenge8/20/2014.
1..
CONTROL VISION Set-up. Step 1 Step 2 Step 3 Step 5 Step 4.
Adding Up In Chunks.
1 © 2004, Cisco Systems, Inc. All rights reserved. CCNA 1 v3.1 Module 2 Networking Fundamentals.
MaK_Full ahead loaded 1 Alarm Page Directory (F11)
1 Processes and Threads Chapter Processes 2.2 Threads 2.3 Interprocess communication 2.4 Classical IPC problems 2.5 Scheduling.
1 10 pt 15 pt 20 pt 25 pt 5 pt 10 pt 15 pt 20 pt 25 pt 5 pt 10 pt 15 pt 20 pt 25 pt 5 pt 10 pt 15 pt 20 pt 25 pt 5 pt 10 pt 15 pt 20 pt 25 pt 5 pt Synthetic.
Dynamic Evolution of Congestion Trees: Analysis and Impact on Switch Architecture P. J. García 1, J. Flich 2, J. Duato 2, I. Johnson 3, F. J. Quiles 1,
Model and Relationships 6 M 1 M M M M M M M M M M M M M M M M
1 hi at no doifpi me be go we of at be do go hi if me no of pi we Inorder Traversal Inorder traversal. n Visit the left subtree. n Visit the node. n Visit.
Analyzing Genes and Genomes
©Brooks/Cole, 2001 Chapter 12 Derived Types-- Enumerated, Structure and Union.
Essential Cell Biology
Clock will move after 1 minute
Intracellular Compartments and Transport
PSSA Preparation.
Essential Cell Biology
Immunobiology: The Immune System in Health & Disease Sixth Edition
Physics for Scientists & Engineers, 3rd Edition
Energy Generation in Mitochondria and Chlorplasts
Delay Analysis and Optimality of Scheduling Policies for Multihop Wireless Networks Gagan Raj Gupta Post-Doctoral Research Associate with the Parallel.
Presentation transcript:

On the Correct Sizing on Meshes Through an Effective Congestion Management Strategy P. J. García 1, J. Flich 2, J. Duato 2, I. Johnson 3, F. J. Quiles 1, F. Naven 3 2 Technical University of Valencia Valencia, Spain 3 Xyratex Havant, UK 1 University of Castilla-La Mancha Albacete, Spain Euro-Par August - 2 SeptemberLisboa, Portugal

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 2 Outline Introduction Congestion and HOL blocking Why HOL blocking affects network sizing? HOL blocking elimination techniques RECN Performance evaluation Conclusions

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 3 Introduction PC Clusters: Alternative to massive parallel computers Current use: –High Performance Computing Systems (HPC) –Internet Servers –Storage Area Network (SANs) Usually based on high-speed interconnection networks High-speed interconnection networks: Myrinet, Infiniband, Quadrics, Advanced Switching… Main features: High bandwidth, Low latencies Additional features: Lossless networks, Flexible topology Network performance may be affected by congestion

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 4 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

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 5 Congestion and HOL Blocking Network contention

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 6 Congestion and HOL Blocking Persistent network contention

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 7 Congestion and HOL Blocking Persistent network contention Flow control

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 8 Congestion and HOL Blocking Persistent network contention Congestion propagates

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 9 Congestion and HOL Blocking Congestion introduces HOL blocking, and this may degrade network performance dramatically 33% HOL 33% 33% 100% 33% 100%

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 10 Why HOL blocking affects network sizing? Network size restricted by: Required system bandwidth: Network offered bandwidth must meet the system traffic conditions Components cost: Recent interconnects (Myrinet, InfiniBand, ASI) are expensive compared to processors Power consumption: As network size increases, higher power consumption, higher heat dissipation Other constraints: Topology, Links per switch, etc. Even if the network is correctly sized, HOL blocking may prevent to reach the expected performance

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 11 Why HOL blocking affects network sizing? Example 1: Reducing cost and consumption A solution is to reduce the number of network components Link utilization increases Low link utilization High link utilization High cost and consumption High congestion and HOL blocking probability

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 12 Why HOL blocking affects network sizing? Example 2: Increasing network bandwidth A solution is to add as network components as necessary Cost, consumption and length of routes increase Greater length of routes Small length of routes Greater HOL blocking probability when congested Low offered bandwidth

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 13 HOL blocking elimination/reduction techniques DAMQs and Virtual Channels not efficient for multihop networks VOQ (Virtual Output Queueing) VOQ at switch level scales but does not eliminate HOL blocking 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

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 14 RECN: Regional Explicit Congestion Notification RECN is a new efficient and scalable congestion management technique Basic ideas: The real problem is not the congestion, but its negative effects (HOL blocking) By eliminating HOL blocking, congestion becomes harmless Non-congested flows do not introduce significant HOL blocking HOL blocking elimination: Packets belonging to congested flows are stored in specific Set Aside Queues (SAQs) Packets belonging to non-congested flows are stored in a common queue Implementation requirements: Deterministic source routing A reduced number of SAQs per port, controlled by a CAM

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 15 How RECN works RECN basic procedure: Congested points are detected in any egress or ingress switch port of the network The routes to detected congested points are progressively notified to ingress and egress ports crossed by congested flows After receiving a notification, a port allocates a SAQ for the detected congested point A packet arriving to a port will be stored in a SAQ if it will pass through the congested point associated to that SAQ A packet arriving at a port will be stored in the common queue if its route does not match any SAQ SAQs can be deallocated, and later allocated for other congested points

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 16 A congestion point forms How RECN Works

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 17 How RECN Works Cold queue fills over a threshold

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 18 How RECN Works

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 19 How RECN Works Internal notification to each input port sending packets to the output port

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 20 How RECN Works

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 21 How RECN Works Input ports allocate a new SAQ for packets addressed to the congested output port

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 22 How RECN Works

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 23 How RECN Works Notification sent when the SAQ fills over a threshold

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 24 How RECN Works

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 25 How RECN Works A new SAQ allocated for the congested port at each output port

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 26 How RECN Works Internal notification when the SAQ fills over A threshold

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 27 How RECN Works The input port allocates A new SAQ

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 28 How RECN Works At the end, the congestion tree builds and is mapped entirely onto SAQs

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 29 Performance Evaluation Two network-sizing scenarios considered: Network cost and consumption reduction: –Network is downsized, keeping constant the number of total system endnodes Network bandwidth increase: –Network size is increased, keeping constant the number of endnodes per switch Evaluation based on simulation results Evaluation metric: Network relative throughput when using: –RECN –VOQ at network level (VOQnet) –VOQ at switch level (VOQsw)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 30 Simulation Model Simulation assumptions: Mesh topologies Deterministic routing (X-Y) 128 KB memories at ingress/egress ports Multiplexed crossbar (BW=12 Gbps) Serial full-duplex pipelined links (BW=8 Gbps) 64-byte packets Credit-based and Xon-Xoff (for SAQs) flow control Maximum of 8 SAQs at ingress/egress ports (RECN)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 31 Network Configurations Five different mesh-based network configurations: Network configuration Mesh topology SwitchesEndnodes per switch #116 x #28 x 8644 #34 x 416 #48 x 8641 #54 x 4161

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 32 Traffic Load Three different synthetic traffic patterns: Normal trafficCongestion tree Traffic case#SourcesDestination#SourcesDestination #1100%Random-- #287.5%Random12.5%hot-spot #375%Random25%hot-spot

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 33 Network Cost and Consumption Reduction Relative throughput: Traffic case endnodes networks (Network configurations 1,2,3) 16 x 16 switches network (Conf. 1) 8 x 8 switches network (Conf. 2) 4 x 4 switches network (Conf. 3)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 34 Network Cost and Consumption Reduction Relative throughput: Traffic case endnodes networks (Network configurations 1,2,3) 16 x 16 switches network (Conf. 1) 8 x 8 switches network (Conf. 2) 4 x 4 switches network (Conf. 3)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 35 Network Cost and Consumption Reduction Relative throughput: Traffic case endnodes networks (Network configurations 1,2,3) 16 x 16 switches network (Conf. 1) 8 x 8 switches network (Conf. 2) 4 x 4 switches network (Conf. 3)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 36 Network Bandwidth Increase Relative throughput: Traffic case 1 1 endnode/switch networks (Network configurations 1,4,5) 16 x 16 switches network (Conf. 1) 8 x 8 switches network (Conf. 4) 4 x 4 switches network (Conf. 5)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 37 Network Bandwidth Increase Relative throughput: Traffic case 2 1 endnode/switch networks (Network configurations 1,4,5) 16 x 16 switches network (Conf. 1) 8 x 8 switches network (Conf. 4) 4 x 4 switches network (Conf. 5)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 38 Network Bandwidth Increase Relative throughput: Traffic case 3 1 endnode/switch networks (Network configurations 1,4,5) 16 x 16 switches network (Conf. 1) 8 x 8 switches network (Conf. 4) 4 x 4 switches network (Conf. 5)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 39 Network Bandwidth Increase Maximum number of SAQs used by RECN: Traffic case 3 1 endnode/switch networks (Network configurations 1,4,5) 16 x 16 switches network (Conf. 1) 8 x 8 switches network (Conf. 4) 4 x 4 switches network (Conf. 5)

Title: On the Correct Sizing on Meshes Through an Effective Congestion Management Strategyy Conference: Euro-Par August-2 SeptemberLisboa, Portugal 40 Conclusions HOL blocking may affect the performance of networks dimensioned with different restrictions We have analyzed the importance of using an efficient HOL blocking elimination strategy We have shown that RECN allows to size the network in any way while keeping network performance at the expected maximum RECN only requires a small number of SAQs for a wide range of network sizes, so it is a scalable strategy

On the Correct Sizing on Meshes Through an Effective Congestion Management Strategy P. J. García 1, J. Flich 2, J. Duato 2, I. Johnson 3, F. J. Quiles 1, F. Naven 3 2 Technical University of Valencia Valencia, Spain 3 Xyratex Havant, UK 1 University of Castilla-La Mancha Albacete, Spain Euro-Par August - 2 SeptemberLisboa, Portugal