Introducing optical switching into the network

Slides:



Advertisements
Similar presentations
EE384Y: Packet Switch Architectures
Advertisements

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
Copyright © 2003 Pearson Education, Inc. Slide 1 Computer Systems Organization & Architecture Chapters 8-12 John D. Carpinelli.
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
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.
1 Optical network CERNET's experience and prospective Xing Li, Congxiao Bao
Congestion Control and Fairness Models Nick Feamster CS 4251 Computer Networking II Spring 2008.
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 ( )
Scalable Routing In Delay Tolerant Networks
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.
FACTORING ax2 + bx + c Think “unfoil” Work down, Show all steps.
Year 6 mental test 5 second questions
Year 6 mental test 10 second questions
1 Understanding Buffer Size Requirements in a Router Thanks to Nick McKeown and John Lockwood for numerous slides.
A Switch-Based Approach to Starvation in Data Centers Alex Shpiner and Isaac Keslassy Department of Electrical Engineering, Technion. Gabi Bracha, Eyal.
REVIEW: Arthropod ID. 1. Name the subphylum. 2. Name the subphylum. 3. Name the order.
(Data and Signals - cont)
1 EE 122: Networks Performance & Modeling Ion Stoica TAs: Junda Liu, DK Moon, David Zats (Materials with thanks.
PP Test Review Sections 6-1 to 6-6
The Weighted Proportional Resource Allocation Milan Vojnović Microsoft Research Joint work with Thành Nguyen Microsoft Research Asia, Beijing, April, 2011.
1 COPYRIGHT © 2011 ALCATEL-LUCENT. ALL RIGHTS RESERVED. On the Capacity of Wireless CSMA/CA Multihop Networks Rafael Laufer and Leonard Kleinrock Bell.
2 |SharePoint Saturday New York City
IP Multicast Information management 2 Groep T Leuven – Information department 2/14 Agenda •Why IP Multicast ? •Multicast fundamentals •Intradomain.
VOORBLAD.
Making Time-stepped Applications Tick in the Cloud Tao Zou, Guozhang Wang, Marcos Vaz Salles*, David Bindel, Alan Demers, Johannes Gehrke, Walker White.
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.
© 2012 National Heart Foundation of Australia. Slide 2.
1 Introduction to Network Layer Lesson 09 NETS2150/2850 School of Information Technologies.
RED-PD: RED with Preferential Dropping Ratul Mahajan Sally Floyd David Wetherall.
Model and Relationships 6 M 1 M M M M M M M M M M M M M M M M
1 Using one or more of your senses to gather information.
Analyzing Genes and Genomes
©Brooks/Cole, 2001 Chapter 12 Derived Types-- Enumerated, Structure and Union.
Essential Cell Biology
Intracellular Compartments and Transport
PSSA Preparation.
Essential Cell Biology
Immunobiology: The Immune System in Health & Disease Sixth Edition
Sizing Router Buffers Guido Appenzeller Isaac Keslassy Nick McKeown Stanford University.
Designing Networks with Little or No Buffers or Can Gulliver Survive in Lilliput? Yashar Ganjali High Performance Networking Group Stanford University.
High Performance All-Optical Networks with Small Buffers Yashar Ganjali High Performance Networking Group Stanford University
High Performance Networking with Little or No Buffers Yashar Ganjali High Performance Networking Group Stanford University
High Performance Networking with Little or No Buffers Yashar Ganjali on behalf of Prof. Nick McKeown High Performance Networking Group Stanford University.
Sizing Router Buffers (Summary)
Sizing Router Buffers Nick McKeown Guido Appenzeller & Isaac Keslassy SNRC Review May 27 th, 2004.
Modeling TCP in Small-Buffer Networks
Reducing the Buffer Size in Backbone Routers Yashar Ganjali High Performance Networking Group Stanford University February 23, 2005
Isaac Keslassy (Technion) Guido Appenzeller & Nick McKeown (Stanford)
Routers with Small Buffers Yashar Ganjali High Performance Networking Group Stanford University
Buffer requirements for TCP: queueing theory & synchronization analysis Gaurav RainaDamon Wischik CambridgeUCL.
CS144 An Introduction to Computer Networks
Congestion models for bursty TCP traffic Damon Wischik + Mark Handley University College London DARPA grant W911NF
How Emerging Optical Technologies will affect the Future Internet NSF Meeting, 5 Dec, 2005 Nick McKeown Stanford University
1 - CS7701 – Fall 2004 Review of: Sizing Router Buffers Paper by: – Guido Appenzeller (Stanford) – Isaac Keslassy (Stanford) – Nick McKeown (Stanford)
Sizing Router Buffers How much packet buffers does a router need? C Router Source Destination 2T The current “Rule of Thumb” A router needs a buffer size:
Buffers: How we fell in love with them, and why we need a divorce Hot Interconnects, Stanford 2004 Nick McKeown High Performance Networking Group Stanford.
Networks with Very Small Buffers Yashar Ganjali, Guido Appenzeller, High Performance Networking Group Prof. Ashish Goel, Prof. Tim Roughgarden, Prof. Nick.
Sachin Katti, CS244 Slides courtesy: Nick McKeown
Understanding Buffer Size Requirements in a Router
Review: Statistical Multiplexing
Routers with Very Small Buffers
Presentation transcript:

Introducing optical switching into the network ECOC 2005, Glasgow Nick McKeown High Performance Networking Group Stanford University nickm@stanford.edu http://www.stanford.edu/~nickm

Widely held assumptions Network religion Bigger buffers are better Widely held assumptions Big buffers (millions of packets) are necessary in Internet routers You can’t build big optical buffers Therefore, you can’t build an all-optical Internet router So how can we use optical switching in the backbone…..? Nick McKeown ECOC 2005

Optical switching in the backbone Approach 1: Don’t use packet switching Dynamic Circuit Switching Approach 2: Packet switching with small buffers Challenge the assumption What we learn in school What we know about the Internet Conclusion: 20 packet buffers might be enough. We’ll consider both… Nick McKeown ECOC 2005

Dynamic circuit switching in the backbone There seem to be some advantages Well-suited to optics Circuit switches are simple “Start with a packet switch and throw most of it away” Higher capacity per unit volume Higher capacity per watt Lower cost per Gb/s There are some disadvantages They are unfashionable Nick McKeown ECOC 2005

Dynamic Circuit Switching (DCS) Some “hurdles” need not be a problem: Routing circuits Signaling Maintaining state Technical hurdles When to create a new circuit? How large to make the circuit? Nick McKeown ECOC 2005

DCS: A circuit per flow aka “TCP Switching” Choose capacity and path/next-hop Simple: 10Gb/s boundary router prototyped on a desktop PC Extreme granularity: Demonstrates feasibility of DCS Nick McKeown ECOC 2005

DCS: Capacity on demand between border routers Rule of thumb Predict the need for capacity by monitoring how quickly new flows are created, rather than waiting for the buffer to fill Nick McKeown ECOC 2005

My conclusion on dynamic circuit switching Compelling to operator: Cost, reliability, management, predictability Scalable with optical circuit switching Users can’t tell the difference Prediction: The backbone will use DCS in 10 years time Nick McKeown ECOC 2005

Optical switching in the backbone Approach 1: Don’t use packet switching Dynamic Circuit Switching Approach 2: Packet switching with small buffers Challenge the assumption What we learn in school What we know about the Internet Conclusion: 20 packet buffers is probably enough. We’ll consider both… Nick McKeown ECOC 2005

Why bigger is not better Network users don’t like buffers Network operators don’t like buffers Router architects don’t like buffers We don’t need big buffers Nick McKeown ECOC 2005

Where did the religion come from? Packet switching is good Long haul links are expensive Statistical multiplexing allows efficient sharing of long haul links Packet switching requires buffers Packet loss is bad Use big buffers Luckily, big electronic buffers are cheap Nick McKeown ECOC 2005

Statistical Multiplexing Observations The bigger the buffer, the lower the packet loss. If the buffer never goes empty, the outgoing line is busy 100% of the time. Nick McKeown ECOC 2005

What we learn in school 1. Queueing Theory r 1 Observations M/M/1 r 1 X Observations Loss rate falls fast with increasing buffer size. Bigger is better. Buffer size Loss rate Nick McKeown ECOC 2005

What we learn in school Moore’s Law: Memory is plentiful and halves in price every 18 months 1Gbit memory: 500k packets for $25 Conclusion Make buffers big Nick McKeown ECOC 2005

Backbone router buffers Source Router Destination C 2T Universally applied rule-of-thumb: A router needs a buffer size: 2T is the two-way propagation delay C is capacity of bottleneck link Context Appears in RFPs and IETF architectural guidelines. Mandated in backbone and edge routers. Villamizar and Song: “High Performance TCP in ANSNET”, CCR, 1994. Known by inventors of TCP [Van Jacobson, 1988] Has major consequences for router design Nick McKeown ECOC 2005

Example 10Gb/s linecard 40Gb/s linecard Rule-of-thumb: Buffer = 1 Million packets 40Gb/s linecard Rule-of-thumb: Buffer = 4 Million packets Requires DRAM, but DRAM too slow Almost unthinkable at 160Gb/s Nick McKeown ECOC 2005

Sizing buffers So where did the rule-of-thumb come from? Answer With , a single TCP flow can keep the bottleneck link busy. Nick McKeown ECOC 2005

TCP Congestion Control Rule for adjusting W If an ACK is received: W ← W+1/W If a packet is lost: W ← W/2 Only W packets may be outstanding Source Dest t Window size Nick McKeown ECOC 2005

TCP Congestion Control Rule for adjusting W If an ACK is received: W ← W+1/W If a packet is lost: W ← W/2 It follows that Only W packets may be outstanding Nick McKeown ECOC 2005

Time evolution of a single TCP flow Time evolution of a single TCP flow through a router. Buffer is 2T*C Time evolution of a single TCP flow through a router. Buffer is < 2T*C Nick McKeown ECOC 2005

Rule-of-thumb Rule-of-thumb makes sense for one flow Typical backbone link has > 20,000 flows Does the rule-of-thumb still hold? Answer: No! Nick McKeown ECOC 2005

If flows were all synchronized t Aggregate window has same dynamics Therefore buffer occupancy has same dynamics Rule-of-thumb still holds. Nick McKeown ECOC 2005

If flows are not synchronized Probability Distribution B Buffer Size Nick McKeown ECOC 2005

Required buffer size Simulation Nick McKeown ECOC 2005

Backbone router buffers Summary The rule of thumb is wrong for a core routers today Required buffer is instead of Validation Theory validated in two small networks (Stanford, University of Wisconsin), two operator networks, and over 10,000 simulations. But more work needed to understand worst-cases… Nick McKeown ECOC 2005

Impact on Router Design 10Gb/s linecard with 200,000 x 56kb/s flows Rule-of-thumb: Buffer = 1 Million packets Becomes: Buffer = 3,000 packets 40Gb/s linecard with 40,000 x 1Mb/s flows Rule-of-thumb: Buffer = 4 Million packets Becomes: Buffer = 25,000 packets Still way too many packets to store optically… Nick McKeown ECOC 2005

What if…? Theory (benign conditions) Practice B Typical OC192 router linecard buffers over 1,000,000 packets M/D/1 Poisson D B Loss independent of link rate, RTT, number of flows, etc. 5 orders of magnitude difference! Can we make traffic look “Poisson-enough” when it arrives to the routers…? Nick McKeown ECOC 2005

Optical routers with very small buffers In collaboration with: Stanford: Yashar Ganjali, Guido Appenzeller, Ashish Goel, Tim Roughgarden LASOR team at DARPA, UCSB, Cisco, JDSU, Calient, Agility Buffer Sizing teams: Frank Kelly (Cambridge), Damon Wischik (UCL), Don Towsley (UMass) Nick McKeown ECOC 2005

Recent work to make buffers smaller 1,000,000 10,000 20 # packets at 10Gb/s Sawtooth Peak-to-trough Smoothing of many sawtooths Non-bursty arrivals Intuition & Proofs Simulated Single TCP Flow Simulated Many TCP Flows Also: Real n/w experiments Simulated Many TCP Flows Just starting: Real n/w experiments Evidence W = window-size of a flow Nick McKeown ECOC 2005

Bursty TCP vs. Spacing packets Bucket drains with a constant rate. Load is 90% for both cases. Nick McKeown ECOC 2005

TCP Reno sends packets in bursts  High drop rate Nick McKeown ECOC 2005

Spacing packets  Much lower drop rate Nick McKeown ECOC 2005

O(log W) Buffers Regular TCP TCP with Spaced Packets Nick McKeown ECOC 2005

O(log W) Buffers With a large ratio between core and access link bandwidth Bottleneck Bandwidth = 1Gb/s Nick McKeown ECOC 2005

O(log W) Buffers Model Assume: link over-provisioned by 1/C, and packets from each flow launched smoothly, or spaced out Window size t Similar results from Cambridge/UCL, UMass and Stanford See papers in: ACM Computer Communications Review, July 2005 Nick McKeown ECOC 2005

Backbone router buffers Summary If the backbone links are 100x faster than the access links, it seems 20 packet buffers will suffice Backbone links will lose 25-30% of their capacity. E.g. a 40Gb/s link will seem like a 28-30Gb/s link For faster flows or access links, use TCP Pacing to space packets Validation Theory and simulation so far Experiments underway… Nick McKeown ECOC 2005

Conclusions Will it be optical DCS or optical packet switching? Technically, both seem feasible Perhaps we shouldn’t care Both are unfashionable I’m hedging my bets… Nick McKeown ECOC 2005