Hardware Evolution in Data Centers
|
|
|
- Aileen McDowell
- 5 years ago
- Views:
Transcription
1
2 Hardware Evolution in Data Centers Trend towards customization Increase work done per dollar (CapEx + OpEx) Paolo Costa Rethinking the Network Stack for Rack-scale Computers 2
3 Hardware Evolution in Data Centers Scale out vs. scale up Many commodity servers rather than few expensive servers Paolo Costa Rethinking the Network Stack for Rack-scale Computers 3
4 Hardware Evolution in Data Centers Custom layout Remove unnecessary components (e.g., GPGPUs, USB ports) Paolo Costa Rethinking the Network Stack for Rack-scale Computers 4
5 Hardware Evolution in Data Centers Integrated fabrics Higher density and bandwidth with lower power consumption Paolo Costa Rethinking the Network Stack for Rack-scale Computers 5
6 Hardware Evolution in Data Centers System-on-Chip (SoC) CPU, IO controllers, NIC/fabric switch on the same die Paolo Costa Rethinking the Network Stack for Rack-scale Computers 6
7 Hardware Evolution in Data Centers Silicon Photonics High-bandwidth / low-latency interconnect (resource disaggregation) Paolo Costa Rethinking the Network Stack for Rack-scale Computers 7
8 Hardware Evolution in Data Centers rack unit (RU) 2 Ru 4-10 Ru Rack-scale intra-server BW << inter-server BW intra-server BW inter-server BW Paolo Costa Rethinking the Network Stack for Rack-scale Computers 8
9 Hardware Evolution in Data Centers Rack-scale Computers The rack is becoming the new unit of computing in data centers Great opportunity to revisit early design assumptions and rethink the software stack and hw/sw co-design rack unit (RU) 2 Ru 4-10 Ru Rack-scale intra-server BW << inter-server BW intra-server BW inter-server BW Paolo Costa Rethinking the Network Stack for Rack-scale Computers 9
10 A First Step: Rethinking the Network Stack Paolo Costa Rethinking the Network Stack for Rack-scale Computers 10
11 A First Step: Rethinking the Network Stack Is it just a faster network or is it fundamentally different? Paolo Costa Rethinking the Network Stack for Rack-scale Computers 11
12 A First Step: Rethinking the Network Stack Many designs are possible but some trends are emerging: Paolo Costa Rethinking the Network Stack for Rack-scale Computers 12
13 A First Step: Rethinking the Network Stack Many designs are possible but some trends are emerging: 1. Distributed Switching Fabric High path diversity Paolo Costa Rethinking the Network Stack for Rack-scale Computers 13
14 A First Step: Rethinking the Network Stack Many designs are possible but some trends are emerging: 1. Distributed Switching Fabric 2. Tight CPU / network integration High path diversity Direct control on network resources Paolo Costa Rethinking the Network Stack for Rack-scale Computers 14
15 A First Step: Rethinking the Network Stack Many designs are possible but some trends are emerging: 1. Distributed Switching Fabric 2. Tight CPU / network integration Research Question: High path diversity How can we take advantage Direct of these control features on network resources to design a new network stack optimized for rack-scale computers? Focus of this work How to route packets (routing)? At what rate should the packets be sent (rate control)? Paolo Costa Rethinking the Network Stack for Rack-scale Computers 15
16 RaSC-Net Design Goals 1. Leverage path diversity and ensure load balance 2. Support arbitrary and dynamic traffic matrixes 3. Achieve low queuing 4. Support custom rate allocation policies priority, deadline-based, tenant-based, resource-based, Paolo Costa Rethinking the Network Stack for Rack-scale Computers 16
17 RaSC-Net Design: Routing A Source Destination B Paolo Costa Rethinking the Network Stack for Rack-scale Computers 17
18 RaSC-Net Design: Routing A Source Destination B Intermediate destination Valiant Load Balancing (VLB) Randomly selects an intermediate destination per each packet Paolo Costa Rethinking the Network Stack for Rack-scale Computers 18
19 RaSC-Net Design: Routing A Source Destination B Intermediate destination Valiant Load Balancing (VLB) Randomly selects an intermediate destination per each packet Paolo Costa Rethinking the Network Stack for Rack-scale Computers 19
20 RaSC-Net Design: Routing A Source Destination B Intermediate destination Valiant Load Balancing (VLB) Randomly selects an intermediate destination per each packet Paolo Costa Rethinking the Network Stack for Rack-scale Computers 20
21 RaSC-Net Design: Routing A Source Destination B Intermediate destination Non-minimal routing Intermediate destinations do not need to lie on the shortest path Paolo Costa Rethinking the Network Stack for Rack-scale Computers 21
22 RaSC-Net Design: Routing A Source Destination B Intermediate destination Non-minimal routing Intermediate destinations do not need to lie on the shortest path Paolo Costa Rethinking the Network Stack for Rack-scale Computers 22
23 RaSC-Net Design: Routing A Source Destination B Intermediate destination Paolo Costa Why does it work? VLB aims to transform any traffic matrix into a uniform one Link load balancing (Goal #1) and independent of traffic matrix (Goal #2) Rethinking the Network Stack for Rack-scale Computers 23
24 RaSC-Net Design: Routing A Source Destination B Intermediate destination VLB Drawback #1: Path Stretch Average path length and link utilization increases Solution: Paolo Costa Locality-aware variants Rethinking the Network of Stack VLB for Rack-scale can Computers be used (weighted choice) 24
25 RaSC-Net Design: Routing A Source Destination B Intermediate destination VLB Drawback #2: Out of order packet arrival Packets need be reordered at the destination (jitter) Paolo Costa Solution: low diameter Rethinking the Network and Stack for minimize Rack-scale Computers queuing delay 25
26 RaSC-Net Design: Routing A C Flow A -> B Source Destination B Intermediate destination Flow-aware Paolo Costa Key observation: VLB enables global visibility Nodes inspect packet headers of forwarded packets They can locally Rethinking the reconstruct Network Stack for Rack-scale the Computers traffic matrix 26
27 RaSC-Net Design: Routing A C Flow A -> B Source Destination B Intermediate destination Flow-aware Paolo Costa Key observation: VLB enables global visibility Nodes inspect packet headers of forwarded packets They can locally Rethinking the reconstruct Network Stack for Rack-scale the Computers traffic matrix 27
28 RaSC-Net Design: Routing A C Flow A -> B Source Destination B Intermediate destination Flow-aware Paolo Costa Key observation: VLB enables global visibility Nodes inspect packet headers of forwarded packets They can locally Rethinking the reconstruct Network Stack for Rack-scale the Computers traffic matrix 28
29 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Rate Allocation Given the knowledge of the topology and the traffic matrix, Paolo Costaeach node can locally Rethinking the derive Network Stack for the Rack-scale correct Computers rate for its flows 29
30 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Rate Allocation Given the knowledge of the topology and the traffic matrix, Paolo Costaeach node can locally Rethinking the derive Network Stack for the Rack-scale correct Computers rate for its flows 30
31 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Rate Allocation Given the knowledge of the topology and the traffic matrix, Paolo Costaeach node can locally Rethinking the derive Network Stack for the Rack-scale correct Computers rate for its flows 31
32 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Rate Allocation Given the knowledge of the topology and the traffic matrix, Paolo Costaeach node can locally Rethinking the derive Network Stack for the Rack-scale correct Computers rate for its flows 32
33 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Rate Adaptation When a packet from a new flow is received, Paolo Costa nodes can Rethinking update the Network Stack their for Rack-scale rates Computers accordingly 33
34 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Key result A traditionally distributed problem (rate control) Paolo Costa becomes Rethinking the Network a local Stack for Rack-scale operation Computers 34
35 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Low queuing (Goal #3) No need to probe the network a la TCP Paolo Costa Rates are chosen Rethinking the Network so Stack as for Rack-scale to minimize Computers queuing 35
36 RaSC-Net Design: Rate Control Flow A -> B Flow C -> D A C Source Flow A -> B Flow C -> D Destination B D Beyond max-min fairness (Goal #4) Since rate allocation is now a local computation, Paolo Costa it is easy to encode Rethinking the Network different Stack for Rack-scale allocation Computers policies 36
37 Something (NOT) to worry about Paolo Costa Rethinking the Network Stack for Rack-scale Computers 37
38 Something (NOT) to worry about 1. Short flows Our protocol needs a few packets to converge We can reserve some spare capacity to absorb short (and newly created) flows Small packets and deterministic routing can also reduce convergence time 2. Computation overhead In general, max/min computation is expensive...but with VLB a flow uses almost all links o Hence, the most bottleneck link is actually the bottleneck for almost all active flows o Approximate results can also be used (utilization vs. computation trade-off) Paolo Costa Rethinking the Network Stack for Rack-scale Computers 38
39 Something (NOT) to worry about 1. Short flows Our protocol needs a few packets to converge We can reserve some spare capacity to absorb short (and newly created) flows Small packets and deterministic routing can also reduce convergence time 2. Computation overhead In general, max/min computation is expensive...but with VLB a flow uses almost all links o Hence, the most bottleneck link is actually the bottleneck for almost all active flows o Approximate results can also be used (utilization vs. computation trade-off) Paolo Costa Rethinking the Network Stack for Rack-scale Computers 39
40 RaSC-Net: Preliminary Results Simulation setup 512-node 3D torus (8 x 8 x 8) 6 10Gbps links / server Permutation matrix with varying number of sources (load) 10-MB flows all starting at the same time Baselines TCP: ECMP routing protocol (single path per flow) + TCP Idealized (unlimited per-flow queues) o Ideal-ShortestPaths: Packet spraying (minimal routing) o Ideal-VLB: Valliant Load Balancing Paolo Costa Rethinking the Network Stack for Rack-scale Computers 40
41 99th Flow Completion Time Flow Completion Time (99 th perc.) Worse TCP Ideal-ShortestPaths Ideal-VLB Better Network load Number of sources increases Paolo Costa Rethinking the Network Stack for Rack-scale Computers 41
42 99th Flow Completion Time Flow Completion Time (99 th perc.) Worse TCP Ideal-ShortestPaths Ideal-VLB RaSC-Net 0.01 Better Network load Number of sources increases Paolo Costa Rethinking the Network Stack for Rack-scale Computers 42
43 99th Flow Completion Time Worse Flow Completion Time (99 th perc.) TCP Ideal-ShortestPaths Ideal-VLB RaSC-Net TCP performance is limited by single path Better 0 Using non-shortest paths increases bandwidth available Network load Number of sources increases RaSC-Net approximates VLB Paolo Costa Rethinking the Network Stack for Rack-scale Computers 43
44 CDF (Flows) Flow Distribution (load = 0.5) TCP Ideal-ShortestPaths Ideal-VLB RaSC-Net Better Flow Completion Time Worse Paolo Costa Rethinking the Network Stack for Rack-scale Computers 44
45 CDF (Flows) Flow Distribution (load = 0.5) achieves lower 1 ShortestPaths median but with a tail Long tail due to single path VLB ensures good Better balance across flows Flow Completion Time TCP Ideal-ShortestPaths Ideal-VLB RaSC-Net The price for fast computation Worse Paolo Costa Rethinking the Network Stack for Rack-scale Computers 45
46 Beyond Rate Control 1. Converged fabric The same fabric is used to carry IP, memory, and storage traffic o Can we design a unified protocol stack (rather than PCI, QPI, SATA, DDR3, )? o How to handle heterogeneous classes of traffic (each with different requirements)? 2. Inter-rack connectivity How to interconnect multiple racks? o Oversubscription? Protocol bridging? 3. Resource management Fast rack fabric blurs the boundaries between local and remote resources o How to assign resources to applications? o How to handle distributed faults? o What s the programming model? Paolo Costa Rethinking the Network Stack for Rack-scale Computers 46
47 Summary Paolo Rack-scale Costa MSR Rethinking Cambridge the Network Stack for Rack-scale Computers 47
From Routing to Traffic Engineering
1 From Routing to Traffic Engineering Robert Soulé Advanced Networking Fall 2016 2 In the beginning B Goal: pair-wise connectivity (get packets from A to B) Approach: configure static rules in routers
Data Center Network Topologies II
Data Center Network Topologies II Hakim Weatherspoon Associate Professor, Dept of Computer cience C 5413: High Performance ystems and Networking April 10, 2017 March 31, 2017 Agenda for semester Project
Optimizing Network Performance in Distributed Machine Learning. Luo Mai Chuntao Hong Paolo Costa
Optimizing Network Performance in Distributed Machine Learning Luo Mai Chuntao Hong Paolo Costa Machine Learning Successful in many fields Online advertisement Spam filtering Fraud detection Image recognition
Mellanox Virtual Modular Switch
WHITE PAPER July 2015 Mellanox Virtual Modular Switch Introduction...1 Considerations for Data Center Aggregation Switching...1 Virtual Modular Switch Architecture - Dual-Tier 40/56/100GbE Aggregation...2
Achieving Efficient Bandwidth Utilization in Wide-Area Networks While Minimizing State Changes
1 Achieving Efficient Bandwidth Utilization in Wide-Area Networks While Minimizing State Changes 2 WAN Traffic Engineering Maintaining private WAN infrastructure is expensive Must balance latency-sensitive
Network-on-Chip Architecture
Multiple Processor Systems(CMPE-655) Network-on-Chip Architecture Performance aspect and Firefly network architecture By Siva Shankar Chandrasekaran and SreeGowri Shankar Agenda (Enhancing performance)
The Network Layer and Routers
The Network Layer and Routers Daniel Zappala CS 460 Computer Networking Brigham Young University 2/18 Network Layer deliver packets from sending host to receiving host must be on every host, router in
TDT Appendix E Interconnection Networks
TDT 4260 Appendix E Interconnection Networks Review Advantages of a snooping coherency protocol? Disadvantages of a snooping coherency protocol? Advantages of a directory coherency protocol? Disadvantages
Optical Packet Switching
Optical Packet Switching DEISNet Gruppo Reti di Telecomunicazioni http://deisnet.deis.unibo.it WDM Optical Network Legacy Networks Edge Systems WDM Links λ 1 λ 2 λ 3 λ 4 Core Nodes 2 1 Wavelength Routing
NaaS Network-as-a-Service in the Cloud
NaaS Network-as-a-Service in the Cloud joint work with Matteo Migliavacca, Peter Pietzuch, and Alexander L. Wolf costa@imperial.ac.uk Motivation Mismatch between app. abstractions & network How the programmers
DeTail Reducing the Tail of Flow Completion Times in Datacenter Networks. David Zats, Tathagata Das, Prashanth Mohan, Dhruba Borthakur, Randy Katz
DeTail Reducing the Tail of Flow Completion Times in Datacenter Networks David Zats, Tathagata Das, Prashanth Mohan, Dhruba Borthakur, Randy Katz 1 A Typical Facebook Page Modern pages have many components
Chapter 7 Slicing and Dicing
1/ 22 Chapter 7 Slicing and Dicing Lasse Harju Tampere University of Technology lasse.harju@tut.fi 2/ 22 Concentrators and Distributors Concentrators Used for combining traffic from several network nodes
Pricing Intra-Datacenter Networks with
Pricing Intra-Datacenter Networks with Over-Committed Bandwidth Guarantee Jian Guo 1, Fangming Liu 1, Tao Wang 1, and John C.S. Lui 2 1 Cloud Datacenter & Green Computing/Communications Research Group
NetSpeed ORION: A New Approach to Design On-chip Interconnects. August 26 th, 2013
NetSpeed ORION: A New Approach to Design On-chip Interconnects August 26 th, 2013 INTERCONNECTS BECOMING INCREASINGLY IMPORTANT Growing number of IP cores Average SoCs today have 100+ IPs Mixing and matching
Forwarding Architecture
Forwarding Architecture Brighten Godfrey CS 538 February 14 2018 slides 2010-2018 by Brighten Godfrey unless otherwise noted Building a fast router Partridge: 50 Gb/sec router A fast IP router well, fast
Interconnection Networks: Topology. Prof. Natalie Enright Jerger
Interconnection Networks: Topology Prof. Natalie Enright Jerger Topology Overview Definition: determines arrangement of channels and nodes in network Analogous to road map Often first step in network design
Routing Algorithms. Review
Routing Algorithms Today s topics: Deterministic, Oblivious Adaptive, & Adaptive models Problems: efficiency livelock deadlock 1 CS6810 Review Network properties are a combination topology topology dependent
Module 17: "Interconnection Networks" Lecture 37: "Introduction to Routers" Interconnection Networks. Fundamentals. Latency and bandwidth
Interconnection Networks Fundamentals Latency and bandwidth Router architecture Coherence protocol and routing [From Chapter 10 of Culler, Singh, Gupta] file:///e /parallel_com_arch/lecture37/37_1.htm[6/13/2012
Interconnection Networks
Lecture 17: Interconnection Networks Parallel Computer Architecture and Programming A comment on web site comments It is okay to make a comment on a slide/topic that has already been commented on. In fact
Routing Domains in Data Centre Networks. Morteza Kheirkhah. Informatics Department University of Sussex. Multi-Service Networks July 2011
Routing Domains in Data Centre Networks Morteza Kheirkhah Informatics Department University of Sussex Multi-Service Networks July 2011 What is a Data Centre? Large-scale Data Centres (DC) consist of tens
Camdoop Exploiting In-network Aggregation for Big Data Applications Paolo Costa
Camdoop Exploiting In-network Aggregation for Big Data Applications costa@imperial.ac.uk joint work with Austin Donnelly, Antony Rowstron, and Greg O Shea (MSR Cambridge) MapReduce Overview Input file
Deadline Guaranteed Service for Multi- Tenant Cloud Storage Guoxin Liu and Haiying Shen
Deadline Guaranteed Service for Multi- Tenant Cloud Storage Guoxin Liu and Haiying Shen Presenter: Haiying Shen Associate professor *Department of Electrical and Computer Engineering, Clemson University,
Utilizing Datacenter Networks: Centralized or Distributed Solutions?
Utilizing Datacenter Networks: Centralized or Distributed Solutions? Costin Raiciu Department of Computer Science University Politehnica of Bucharest We ve gotten used to great applications Enabling Such
DevoFlow: Scaling Flow Management for High-Performance Networks
DevoFlow: Scaling Flow Management for High-Performance Networks Andy Curtis Jeff Mogul Jean Tourrilhes Praveen Yalagandula Puneet Sharma Sujata Banerjee Software-defined networking Software-defined networking
4. Networks. in parallel computers. Advances in Computer Architecture
4. Networks in parallel computers Advances in Computer Architecture System architectures for parallel computers Control organization Single Instruction stream Multiple Data stream (SIMD) All processors
PUSHING THE LIMITS, A PERSPECTIVE ON ROUTER ARCHITECTURE CHALLENGES
PUSHING THE LIMITS, A PERSPECTIVE ON ROUTER ARCHITECTURE CHALLENGES Greg Hankins APRICOT 2012 2012 Brocade Communications Systems, Inc. 2012/02/28 Lookup Capacity and Forwarding
Episode 5. Scheduling and Traffic Management
Episode 5. Scheduling and Traffic Management Part 3 Baochun Li Department of Electrical and Computer Engineering University of Toronto Outline What is scheduling? Why do we need it? Requirements of a scheduling
Priority Traffic CSCD 433/533. Advanced Networks Spring Lecture 21 Congestion Control and Queuing Strategies
CSCD 433/533 Priority Traffic Advanced Networks Spring 2016 Lecture 21 Congestion Control and Queuing Strategies 1 Topics Congestion Control and Resource Allocation Flows Types of Mechanisms Evaluation
Chapter 1. Introduction
Chapter 1 Introduction In a packet-switched network, packets are buffered when they cannot be processed or transmitted at the rate they arrive. There are three main reasons that a router, with generic
Topologies. Maurizio Palesi. Maurizio Palesi 1
Topologies Maurizio Palesi Maurizio Palesi 1 Network Topology Static arrangement of channels and nodes in an interconnection network The roads over which packets travel Topology chosen based on cost and
Interconnection Networks: Routing. Prof. Natalie Enright Jerger
Interconnection Networks: Routing Prof. Natalie Enright Jerger Routing Overview Discussion of topologies assumed ideal routing In practice Routing algorithms are not ideal Goal: distribute traffic evenly
DevoFlow: Scaling Flow Management for High Performance Networks
DevoFlow: Scaling Flow Management for High Performance Networks SDN Seminar David Sidler 08.04.2016 1 Smart, handles everything Controller Control plane Data plane Dump, forward based on rules Existing
Optical Interconnection Networks in Data Centers: Recent Trends and Future Challenges
Optical Interconnection Networks in Data Centers: Recent Trends and Future Challenges Speaker: Lin Wang Research Advisor: Biswanath Mukherjee Kachris C, Kanonakis K, Tomkos I. Optical interconnection networks
Scaling to Petaflop. Ola Torudbakken Distinguished Engineer. Sun Microsystems, Inc
Scaling to Petaflop Ola Torudbakken Distinguished Engineer Sun Microsystems, Inc HPC Market growth is strong CAGR increased from 9.2% (2006) to 15.5% (2007) Market in 2007 doubled from 2003 (Source: IDC
Basic Switch Organization
NOC Routing 1 Basic Switch Organization 2 Basic Switch Organization Link Controller Used for coordinating the flow of messages across the physical link of two adjacent switches 3 Basic Switch Organization
XORs in the Air: Practical Wireless Network Coding
XORs in the Air: Practical Wireless Network Coding S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard, J. Crowcroft MIT & University of Cambridge Can we use 3 transmissions to send traffic? 1 2 4 3 Can we
Introduction to Segment Routing
Segment Routing (SR) is a flexible, scalable way of doing source routing. Overview of Segment Routing, page 1 How Segment Routing Works, page 2 Examples for Segment Routing, page 3 Benefits of Segment
Introduction. Network Architecture Requirements of Data Centers in the Cloud Computing Era
Massimiliano Sbaraglia Network Engineer Introduction In the cloud computing era, distributed architecture is used to handle operations of mass data, such as the storage, mining, querying, and searching
A Scalable, Commodity Data Center Network Architecture
A Scalable, Commodity Data Center Network Architecture B Y M O H A M M A D A L - F A R E S A L E X A N D E R L O U K I S S A S A M I N V A H D A T P R E S E N T E D B Y N A N X I C H E N M A Y. 5, 2 0
CS6453. Data-Intensive Systems: Rachit Agarwal. Technology trends, Emerging challenges & opportuni=es
CS6453 Data-Intensive Systems: Technology trends, Emerging challenges & opportuni=es Rachit Agarwal Slides based on: many many discussions with Ion Stoica, his class, and many industry folks Servers Typical
CS 498 Hot Topics in High Performance Computing. Networks and Fault Tolerance. 9. Routing and Flow Control
CS 498 Hot Topics in High Performance Computing Networks and Fault Tolerance 9. Routing and Flow Control Intro What did we learn in the last lecture Topology metrics Including minimum diameter of directed
Basics (cont.) Characteristics of data communication technologies OSI-Model
48 Basics (cont.) Characteristics of data communication technologies OSI-Model Topologies Packet switching / Circuit switching Medium Access Control (MAC) mechanisms Coding Quality of Service (QoS) 49
DCRoute: Speeding up Inter-Datacenter Traffic Allocation while Guaranteeing Deadlines
DCRoute: Speeding up Inter-Datacenter Traffic Allocation while Guaranteeing Deadlines Mohammad Noormohammadpour, Cauligi S. Raghavendra Ming Hsieh Department of Electrical Engineering University of Southern
Network-on-chip (NOC) Topologies
Network-on-chip (NOC) Topologies 1 Network Topology Static arrangement of channels and nodes in an interconnection network The roads over which packets travel Topology chosen based on cost and performance
SPARTA: Scalable Per-Address RouTing Architecture
SPARTA: Scalable Per-Address RouTing Architecture John Carter Data Center Networking IBM Research - Austin IBM Research Science & Technology IBM Research activities related to SDN / OpenFlow IBM Research
A Low Latency Solution Stack for High Frequency Trading. High-Frequency Trading. Solution. White Paper
A Low Latency Solution Stack for High Frequency Trading White Paper High-Frequency Trading High-frequency trading has gained a strong foothold in financial markets, driven by several factors including
Slim Fly: A Cost Effective Low-Diameter Network Topology
TORSTEN HOEFLER, MACIEJ BESTA Slim Fly: A Cost Effective Low-Diameter Network Topology Images belong to their creator! NETWORKS, LIMITS, AND DESIGN SPACE Networks cost 25-30% of a large supercomputer Hard
Cisco Extensible Network Controller
Data Sheet Cisco Extensible Network Controller Product Overview Today s resource intensive applications are making the network traffic grow exponentially putting high demands on the existing network. Companies
Performance Analysis of Storage-Based Routing for Circuit-Switched Networks [1]
![Performance Analysis of Storage-Based Routing for Circuit-Switched Networks [1]](/thumbs/89/97802209.jpg "Performance Analysis of Storage-Based Routing for Circuit-Switched Networks [1]")
Performance Analysis of Storage-Based Routing for Circuit-Switched Networks [1] Presenter: Yongcheng (Jeremy) Li PhD student, School of Electronic and Information Engineering, Soochow University, China
Data Center Networks and Switching and Queueing and Covert Timing Channels
Data Center Networks and Switching and Queueing and Covert Timing Channels Hakim Weatherspoon Associate Professor, Dept of Computer Science CS 5413: High Performance Computing and Networking March 10,
Reducing Network Tiers Flattening the Network. Kevin Ryan Director Data Center Solutions
Reducing Tiers Flattening the Kevin Ryan Director Data Center Solutions www.extremenetworks.com Data Center Trends The New Computer Data center capacity, not server capacity, is the new metric Consolidation
IV. PACKET SWITCH ARCHITECTURES
IV. PACKET SWITCH ARCHITECTURES (a) General Concept - as packet arrives at switch, destination (and possibly source) field in packet header is used as index into routing tables specifying next switch in
Lecture 21: Congestion Control" CSE 123: Computer Networks Alex C. Snoeren
Lecture 21: Congestion Control" CSE 123: Computer Networks Alex C. Snoeren Lecture 21 Overview" How fast should a sending host transmit data? Not to fast, not to slow, just right Should not be faster than
Chapter 4. Routers with Tiny Buffers: Experiments. 4.1 Testbed experiments Setup
Chapter 4 Routers with Tiny Buffers: Experiments This chapter describes two sets of experiments with tiny buffers in networks: one in a testbed and the other in a real network over the Internet2 1 backbone.
Getting Real Performance from a Virtualized CCAP
Getting Real Performance from a Virtualized CCAP A Technical Paper prepared for SCTE/ISBE by Mark Szczesniak Software Architect Casa Systems, Inc. 100 Old River Road Andover, MA, 01810 978-688-6706 mark.szczesniak@casa-systems.com
Deploying Data Center Switching Solutions
Deploying Data Center Switching Solutions Choose the Best Fit for Your Use Case 1 Table of Contents Executive Summary... 3 Introduction... 3 Multivector Scaling... 3 Low On-Chip Memory ASIC Platforms...4
DARD: A Practical Distributed Adaptive Routing Architecture for Datacenter Networks
DARD: A Practical Distributed Adaptive Routing Architecture for Datacenter Networks Xin Wu and Xiaowei Yang Duke-CS-TR-20-0 {xinwu, xwy}@cs.duke.edu Abstract Datacenter networks typically have multiple
Enabling High Performance Data Centre Solutions and Cloud Services Through Novel Optical DC Architectures. Dimitra Simeonidou
Enabling High Performance Data Centre Solutions and Cloud Services Through Novel Optical DC Architectures Dimitra Simeonidou Challenges and Drivers for DC Evolution Data centres are growing in size and
Routing Algorithm. How do I know where a packet should go? Topology does NOT determine routing (e.g., many paths through torus)
Routing Algorithm How do I know where a packet should go? Topology does NOT determine routing (e.g., many paths through torus) Many routing algorithms exist 1) Arithmetic 2) Source-based 3) Table lookup
Memory: Past, Present and Future Trends Paolo Faraboschi
Memory: Past, Present and Future Trends Paolo Faraboschi Fellow, Hewlett Packard Labs Systems Research Lab Quiz ( Excerpt from Intel Developer Forum Keynote 2000 ) ANDREW GROVE: is there a role for more
Topology basics. Constraints and measures. Butterfly networks.
EE48: Advanced Computer Organization Lecture # Interconnection Networks Architecture and Design Stanford University Topology basics. Constraints and measures. Butterfly networks. Lecture #: Monday, 7 April
MUD: Send me your top 1 3 questions on this lecture
Administrivia Review 1 due tomorrow Email your reviews to me Office hours on Thursdays 10 12 MUD: Send me your top 1 3 questions on this lecture Guest lectures next week by Prof. Richard Martin Class slides
AGILE OPTICAL NETWORK (AON) CORE NETWORK USE CASES
AGILE OPTICAL NETWORK (AON) CORE NETWORK USE CASES PAOLO FOGLIATA - EMEA IP Transport, Director Optical Networking Dec 3, 2014 1830 PSS MARKET MOMENTUM 100G DEPLOYMENTS ACCELERATE 300+/WEEK 400 Manufactured
Performance of Multihop Communications Using Logical Topologies on Optical Torus Networks
Performance of Multihop Communications Using Logical Topologies on Optical Torus Networks X. Yuan, R. Melhem and R. Gupta Department of Computer Science University of Pittsburgh Pittsburgh, PA 156 fxyuan,
Achieving Lightweight Multicast in Asynchronous Networks-on-Chip Using Local Speculation
Achieving Lightweight Multicast in Asynchronous Networks-on-Chip Using Local Speculation Kshitij Bhardwaj Dept. of Computer Science Columbia University Steven M. Nowick 2016 ACM/IEEE Design Automation
Fast-Response Multipath Routing Policy for High-Speed Interconnection Networks
HPI-DC 09 Fast-Response Multipath Routing Policy for High-Speed Interconnection Networks Diego Lugones, Daniel Franco, and Emilio Luque Leonardo Fialho Cluster 09 August 31 New Orleans, USA Outline Scope
CellSDN: Software-Defined Cellular Core networks
CellSDN: Software-Defined Cellular Core networks Xin Jin Princeton University Joint work with Li Erran Li, Laurent Vanbever, and Jennifer Rexford Cellular Core Network Architecture Base Station User Equipment
Managed IP Services from Dial Access to Gigabit Routers
Managed IP Services from Dial Access to Gigabit Routers Technical barriers and Future trends for IP Differentiated Services Grenville Armitage, PhD Member of Technical Staff High Speed Networks Research,
Lecture 28: Networks & Interconnect Architectural Issues Professor Randy H. Katz Computer Science 252 Spring 1996
Lecture 28: Networks & Interconnect Architectural Issues Professor Randy H. Katz Computer Science 252 Spring 1996 RHK.S96 1 Review: ABCs of Networks Starting Point: Send bits between 2 computers Queue
Lecture 2: Topology - I
ECE 8823 A / CS 8803 - ICN Interconnection Networks Spring 2017 http://tusharkrishna.ece.gatech.edu/teaching/icn_s17/ Lecture 2: Topology - I Tushar Krishna Assistant Professor School of Electrical and
Unit 2 Packet Switching Networks - II
Unit 2 Packet Switching Networks - II Dijkstra Algorithm: Finding shortest path Algorithm for finding shortest paths N: set of nodes for which shortest path already found Initialization: (Start with source
DISTRIBUTED EMBEDDED ARCHITECTURES
DISTRIBUTED EMBEDDED ARCHITECTURES A distributed embedded system can be organized in many different ways, but its basic units are the Processing Elements (PE) and the network as illustrated in Figure.
Crossbar - example. Crossbar. Crossbar. Combination: Time-space switching. Simple space-division switch Crosspoints can be turned on or off
Crossbar Crossbar - example Simple space-division switch Crosspoints can be turned on or off i n p u t s sessions: (,) (,) (,) (,) outputs Crossbar Advantages: simple to implement simple control flexible
Revisiting Network Support for RDMA
Revisiting Network Support for RDMA Radhika Mittal 1, Alex Shpiner 3, Aurojit Panda 1, Eitan Zahavi 3, Arvind Krishnamurthy 2, Sylvia Ratnasamy 1, Scott Shenker 1 (1: UC Berkeley, 2: Univ. of Washington,
Investigating the Use of Synchronized Clocks in TCP Congestion Control
Investigating the Use of Synchronized Clocks in TCP Congestion Control Michele Weigle (UNC-CH) November 16-17, 2001 Univ. of Maryland Symposium The Problem TCP Reno congestion control reacts only to packet
Lecture 13: Traffic Engineering
Lecture 13: Traffic Engineering CSE 222A: Computer Communication Networks Alex C. Snoeren Thanks: Mike Freedman, Nick Feamster Lecture 13 Overview Evolution of routing in the ARPAnet Today s TE: Adjusting
Scalable Distributed Training with Parameter Hub: a whirlwind tour
Scalable Distributed Training with Parameter Hub: a whirlwind tour TVM Stack Optimization High-Level Differentiable IR Tensor Expression IR AutoTVM LLVM, CUDA, Metal VTA AutoVTA Edge FPGA Cloud FPGA ASIC
PowerEdge NUMA Configurations with AMD EPYC Processors
PowerEdge Product Group Direct from Development PowerEdge NUMA Configurations with AMD EPYC Processors Tech Note by: Jose Grande SUMMARY With the introduction of AMD s EPYC (Naples) x86 Server CPUs featuring
Overview. TCP & router queuing Computer Networking. TCP details. Workloads. TCP Performance. TCP Performance. Lecture 10 TCP & Routers
Overview 15-441 Computer Networking TCP & router queuing Lecture 10 TCP & Routers TCP details Workloads Lecture 10: 09-30-2002 2 TCP Performance TCP Performance Can TCP saturate a link? Congestion control
CSE 123: Computer Networks Alex C. Snoeren. HW 2 due Thursday 10/21!
CSE 123: Computer Networks Alex C. Snoeren HW 2 due Thursday 10/21! Finishing up media access Contention-free methods (rings) Moving beyond one wire Link technologies have limits on physical distance Also
The GLIMPS Terabit Packet Switching Engine
February 2002 The GLIMPS Terabit Packet Switching Engine I. Elhanany, O. Beeri Terabit Packet Switching Challenges The ever-growing demand for additional bandwidth reflects on the increasing capacity requirements
An FPGA-Based Optical IOH Architecture for Embedded System
An FPGA-Based Optical IOH Architecture for Embedded System Saravana.S Assistant Professor, Bharath University, Chennai 600073, India Abstract Data traffic has tremendously increased and is still increasing
Modeling UGAL routing on the Dragonfly topology
Modeling UGAL routing on the Dragonfly topology Md Atiqul Mollah, Peyman Faizian, Md Shafayat Rahman, Xin Yuan Florida State University Scott Pakin, Michael Lang Los Alamos National Laboratory Motivation
Contention-based Congestion Management in Large-Scale Networks
Contention-based Congestion Management in Large-Scale Networks Gwangsun Kim, Changhyun Kim, Jiyun Jeong, Mike Parker, John Kim KAIST Intel Corp. {gskim, nangigs, cjy9037, jjk12}@kaist.ac.kr mike.a.parker@intel.com
Switching and Forwarding Reading: Chapter 3 1/30/14 1
Switching and Forwarding Reading: Chapter 3 1/30/14 1 Switching and Forwarding Next Problem: Enable communication between hosts that are not directly connected Fundamental Problem of the Internet or any
Next Generation Networks MultiService Network Design. Dr. Ben Tang
Next Generation Networks MultiService Network Design Dr. Ben Tang April 23, 2008 Operators face difficult network planning and design questions Can your current planning methods handle the complexity of
Sort vs. Hash Join Revisited for Near-Memory Execution. Nooshin Mirzadeh, Onur Kocberber, Babak Falsafi, Boris Grot
Sort vs. Hash Join Revisited for Near-Memory Execution Nooshin Mirzadeh, Onur Kocberber, Babak Falsafi, Boris Grot 1 Near-Memory Processing (NMP) Emerging technology Stacked memory: A logic die w/ a stack
Combining Arm & RISC-V in Heterogeneous Designs
Combining Arm & RISC-V in Heterogeneous Designs Gajinder Panesar, CTO, UltraSoC gajinder.panesar@ultrasoc.com RISC-V Summit 3 5 December 2018 Santa Clara, USA Problem statement Deterministic multi-core
OCP Engineering Workshop - Telco
OCP Engineering Workshop - Telco Low Latency Mobile Edge Computing Trevor Hiatt Product Management, IDT IDT Company Overview Founded 1980 Workforce Approximately 1,800 employees Headquarters San Jose,
Network Design Considerations for Grid Computing
Network Design Considerations for Grid Computing Engineering Systems How Bandwidth, Latency, and Packet Size Impact Grid Job Performance by Erik Burrows, Engineering Systems Analyst, Principal, Broadcom
Advanced Computer Networks. End Host Optimization
Oriana Riva, Department of Computer Science ETH Zürich 263 3501 00 End Host Optimization Patrick Stuedi Spring Semester 2017 1 Today End-host optimizations: NUMA-aware networking Kernel-bypass Remote Direct
How Emerging Optical Technologies will affect the Future Internet
How Emerging Optical Technologies will affect the Future Internet NSF Meeting, 5 Dec, 2005 Nick McKeown Stanford University nickm@stanford.edu http://www.stanford.edu/~nickm Emerged (and deployed) Optical
A common issue that affects the QoS of packetized audio is jitter. Voice data requires a constant packet interarrival rate at receivers to convert
A common issue that affects the QoS of packetized audio is jitter. Voice data requires a constant packet interarrival rate at receivers to convert data into a proper analog signal for playback. The variations
Adaptive Video Multicasting
Adaptive Video Multicasting Presenters: Roman Glistvain, Bahman Eksiri, and Lan Nguyen 1 Outline Approaches to Adaptive Video Multicasting Single rate multicast Simulcast Active Agents Layered multicasting
ECE 4750 Computer Architecture, Fall 2017 T06 Fundamental Network Concepts
ECE 4750 Computer Architecture, Fall 2017 T06 Fundamental Network Concepts School of Electrical and Computer Engineering Cornell University revision: 2017-10-17-12-26 1 Network/Roadway Analogy 3 1.1. Running
Datacenter Wide- area Enterprise
Datacenter Wide- area Enterprise Client LOAD- BALANCER Can t choose path : ( Servers Outline and goals A new architecture for distributed load-balancing joint (server, path) selection Demonstrate a nation-wide
DATA CENTER FABRIC COOKBOOK
Do It Yourself! DATA CENTER FABRIC COOKBOOK How to prepare something new from well known ingredients Emil Gągała WHAT DOES AN IDEAL FABRIC LOOK LIKE? 2 Copyright 2011 Juniper Networks, Inc. www.juniper.net
CSC 4900 Computer Networks: Network Layer
CSC 4900 Computer Networks: Network Layer Professor Henry Carter Fall 2017 Villanova University Department of Computing Sciences Review What is AIMD? When do we use it? What is the steady state profile
Data Centers. Tom Anderson
Data Centers Tom Anderson Transport Clarification RPC messages can be arbitrary size Ex: ok to send a tree or a hash table Can require more than one packet sent/received We assume messages can be dropped,
Single Root I/O Virtualization (SR-IOV) and iscsi Uncompromised Performance for Virtual Server Environments Leonid Grossman Exar Corporation
Single Root I/O Virtualization (SR-IOV) and iscsi Uncompromised Performance for Virtual Server Environments Leonid Grossman Exar Corporation Introduction to Exar iscsi project and related datacenter trends