Congestion Notification Mechanisms in 802 networks Manoj Wadekar IEEE Interim Meeting January 12, 2005.

Slides:



Advertisements
Similar presentations
Ethernet Congestion Management Brad Booth, Intel September 2004.
Advertisements

Data Link Layer B. Konkoth. PDU  Protocol Data Unit  A unit of data which is specified in a protocol of a given layer  Layer 5, 6, 7 – Data  Layer.
Guidelines for Adding Congestion Notification to Protocols that Encapsulate IP draft-briscoe-tsvwg-ecn-encap-guidelines-00 Bob Briscoe IETF-80 Mar 2011.
© 2007 Cisco Systems, Inc. All rights reserved.Cisco Public 1 Addressing the Network – IPv4 Network Fundamentals – Chapter 6.
Cs/ee 143 Communication Networks Chapter 6 Internetworking Text: Walrand & Parekh, 2010 Steven Low CMS, EE, Caltech.
VIA and Its Extension To TCP/IP Network Yingping Lu Based on Paper “Queue Pair IP, …” by Philip Buonadonna.
1 Congestion Control. Transport Layer3-2 Principles of Congestion Control Congestion: r informally: “too many sources sending too much data too fast for.
1 Internet Networking Spring 2003 Tutorial 11 Explicit Congestion Notification (RFC 3168) Limited Transmit (RFC 3042)
1 Version 3 Module 8 Ethernet Switching. 2 Version 3 Ethernet Switching Ethernet is a shared media –One node can transmit data at a time More nodes increases.
OSI Model.
1 Internet Networking Spring 2003 Tutorial 11 Explicit Congestion Notification (RFC 3168)
Introduction to Management Information Systems Chapter 5 Data Communications and Internet Technology HTM 304 Fall 07.
1 Spring Semester 2007, Dept. of Computer Science, Technion Internet Networking recitation #8 Explicit Congestion Notification (RFC 3168) Limited Transmit.
1 Interconnecting LAN segments Repeaters Hubs Bridges Switches.
CS335 Networking & Network Administration Tuesday, April 20, 2010.
COS 461: Computer Networks
Networking and Internetworking Devices Networks and Protocols Prepared by: TGK First Prepared on: Last Modified on: Quality checked by: Copyright 2009.
Institute of Technology, Sligo Dept of Computing Semester 3, version Semester 3 Chapter 3 VLANs.
(part 3).  Switches, also known as switching hubs, have become an increasingly important part of our networking today, because when working with hubs,
IWARP Ethernet Key to Driving Ethernet into the Future Brian Hausauer Chief Architect NetEffect, Inc.
Networking Components Chad Benedict – LTEC
Lawrence G. Roberts CEO Anagran September 2005 Advances Toward Economic and Efficient Terabit LANs and WANs.
We will be covering VLANs this week. In addition we will do a practical involving setting up a router and how to create a VLAN.
© 2007 Cisco Systems, Inc. All rights reserved.Cisco Public 1 Addressing the Network – IPv4 Network Fundamentals – Chapter 6.
CECS 5460 – Assignment 3 Stacey VanderHeiden Güney.
Network Redundancy Multiple paths may exist between systems. Redundancy is not a requirement of a packet switching network. Redundancy was part of the.
Chapter 4: Managing LAN Traffic
Chapter 1 Overview Review Overview of demonstration network
Lecture 2 TCP/IP Protocol Suite Reference: TCP/IP Protocol Suite, 4 th Edition (chapter 2) 1.
ACM 511 Chapter 2. Communication Communicating the Messages The best approach is to divide the data into smaller, more manageable pieces to send over.
Towards a Common Communication Infrastructure for Clusters and Grids Darius Buntinas Argonne National Laboratory.
CS3502: Data and Computer Networks Local Area Networks - 4 Bridges / LAN internetworks.
Module 4: Designing Routing and Switching Requirements.
1 © 2003, Cisco Systems, Inc. All rights reserved. CCNA 3 v3.0 Module 4 Switching Concepts.
CMPT 471 Networking II Address Resolution IPv4 ARP RARP 1© Janice Regan, 2012.
The NE010 iWARP Adapter Gary Montry Senior Scientist
Communication Networks Fourth Meeting. Types of Networks  What is a circuit network?  Two people are connected and allocated them their own physical.
Chapter 8: Virtual LAN (VLAN)
By: Aleksandr Movsesyan Advisor: Hugh Smith. OSI Model.
1 Countering DoS Through Filtering Omar Bashir Communications Enabling Technologies
1 © 2003, Cisco Systems, Inc. All rights reserved. CCNA 3 v3.0 Module 4 Switching Concepts.
1 Congestion Control Computer Networks. 2 Where are we?
Congestion Management Study Group1 September 2004 PAR Title Information technology -- Telecommunications and information exchange between systems -- Local.
1 © 2003, Cisco Systems, Inc. All rights reserved. CCNA 3 v3.0 Module 4 Switching Concepts.
First, by sending smaller individual pieces from source to destination, many different conversations can be interleaved on the network. The process.
Chapter 3 - VLANs. VLANs Logical grouping of devices or users Configuration done at switch via software Not standardized – proprietary software from vendor.
Ethernet. Ethernet standards milestones 1973: Ethernet Invented 1983: 10Mbps Ethernet 1985: 10Mbps Repeater 1990: 10BASE-T 1995: 100Mbps Ethernet 1998:
Chapter 3: Network Protocols and Communications
1 Connectivity with ARP and RARP. 2 There needs to be a mapping between the layer 2 and layer 3 addresses (i.e. IP to Ethernet). Mapping should be dynamic.
Internet Protocol Storage Area Networks (IP SAN)
Networking Protocols John R. Durrett ISQS 6343 #1.
Data Communications and Computer Networks Chapter 4 CS 3830 Lecture 19 Omar Meqdadi Department of Computer Science and Software Engineering University.
Sockets Direct Protocol for Hybrid Network Stacks: A Case Study with iWARP over 10G Ethernet P. Balaji, S. Bhagvat, R. Thakur and D. K. Panda, Mathematics.
Data Communication 1 Frame Relay n X.25 l Provides extensive error checking and flow control l station-to-station checking at the data link layer l Error.
Schutzvermerk nach DIN 34 beachten Ethernet Deterministics.
1 CCNA 3 v3.1 Module 4 Switching Concepts Claes Larsen, CCAI.
Ethernet Congestion Management Brad Booth, Intel September 2004.
© 2007 EMC Corporation. All rights reserved. Internet Protocol Storage Area Networks (IP SAN) Module 3.4.
Congestion Notification Process for Real-Time Traffic draft-babiarz-tsvwg-rtecn-04.txt Jozef Babiarz Kwok Ho Chan
1 Three ways to (ab)use Multipath Congestion Control Costin Raiciu University Politehnica of Bucharest.
Network Processing Systems Design
Link Layer 5.1 Introduction and services
Internet Networking recitation #9
A quick intro to networking
Chapter 4 Data Link Layer Switching
ADDRESSING Before you can send a message, you must know the destination address. It is extremely important to understand that each computer has several.
Internet Networking recitation #10
Chapter 3 VLANs Chaffee County Academy
DetNet Architecture Updates
Presentation transcript:

Congestion Notification Mechanisms in 802 networks Manoj Wadekar IEEE Interim Meeting January 12, 2005

Page 2 Intel Corp. Agenda Market Potential Requirements and Scope Congestion Notification mechanisms Proposal for L2 mechanism – L2-CI Summary

Page 3 Intel Corp. Summary of request In order to enable accelerated deployment of Ethernet into emerging limited-topology applications (clustering, backplanes, storage, data centers, etc.), IEEE should specify a standard mechanism for MAC Clients to provide congestion information to L2 edge devices, using wadekar_1_0501.pdf as a basis

Page 4 Intel Corp. Congestion Control Elements Detection  Could be an AQM like RED – Does not need to be specified by IEEE 802 Notification  Need a standard way to notify congestion between L2 devices Request to IEEE to consider Action  Rate control/reduction done by source in response to congestion notification  Left to ULPs (L3 and above) e.g. TCP IETF Domain

Market Potential

Page 6 Intel Corp. Market Opportunities for Ethernet Extend Ethernet Reach by improving congestion management capabilities IT Perceptions about Ethernet: “Ethernet not adequate for low latency apps” “Ethernet frame loss is inefficient for storage” Market Opportunity Clustering & Grid computing (RDMA, iWARP) Storage (iSCSI) Telco Backplanes

Page 7 Intel Corp. Emerging Blade Usage Models Front End (FE) Edge Servers Mid Tier (MT) Application Servers Back End (BE) Data Servers NAS DAS SAN DAS Blades are increasingly being deployed in BE & MT applications Ethernet is the default fabric of choice for LAN  In addition to Ethernet, Blades use Fiber Channel and Infiniband® for supporting Storage and Inter-processor communication traffic today Ethernet Blades are a growing piece of Telco pie ~ 26% of Telco servers by ‘07 – In-Stat/MDR

Requirements and Scope

Page 9 Intel Corp. CM Requirements for Datacenter Address IT perceptions:  “Ethernet not adequate for low latency apps”  “Ethernet frame loss is inefficient for storage” Improve Ethernet Congestion Management capabilities that will:  Reduce frame loss significantly  Reduce end-to-end latency and latency jitter  Achieve above without compromising throughput Address needs of Short Range Networks  Backplanes  Clusters BUT “Do No harm” if enabled in other topologies

Page 10 Intel Corp. CMSG Discussions - Recap Existing Link level mechanisms for congestion control do not improve network throughput  Head of line blocking  Congestion spreading  Increase jitter for high-priority traffic  Sacrifices throughput for avoiding frame loss Congestion control can be done at data source that is causing congestion  However, congestion happens somewhere else (bridges, destination nodes etc.) Congested devices need to provide information finally to source  Data sources can respond by reducing traffic into congested paths

Page 11 Intel Corp. Applicability of CN from Bridges Congestion Management is achieved by:  Bridges providing congestion information  Data Sources (ULP) providing Rate Control mechanisms Remaining presentation focuses on Ethernet (802.3) networks However, enhancements may be viable for other networks as well  , etc.

Page 12 Intel Corp. Congestion Control Elements Detection  Could be an AQM like RED – Does not need to be specified by IEEE 802 Notification  Need a standard way to notify congestion between L2 devices Request to IEEE to consider Action  Rate control/reduction done by source in response to congestion notification  Left to ULPs (L3 and above) e.g. TCP IETF Domain

Congestion Notification Mechanisms

Page 14 Intel Corp. Congestion Indication mechanisms Packet Marking (triggered by congestion event)  Forward Marking of the packet experiencing congestion Leave it to upper protocol for getting information back to the source  Or Backward Marking of packets going to congestion source Which source (L2, Upper Protocol, what granularity)? Control Message  Send control packet to congestion source triggered by congestion Which source? Granularity - L2, Upper Protocol, Socket,?? Should be in fast-path  Periodic Control messages carrying congestion information

Page 15 Intel Corp. More discussion on Backward Notification Faster turnaround, support for asymmetric traffic sources (e.g. non-TCP flows) Backward Notification creates traffic in congested networks  Can argue that transient congestions may not affect same paths simultaneously How to define granularity  Is L2 information sufficient?

Page 16 Intel Corp. L3 Marking Mechanisms : IP-CE IP – CE (Congestion Experienced)  IP-CE marking by routers or L2+ Switches when congestion is experienced Pros:  Will provide ECN capability within L2 Subnet  No change required in end-station implementations Cons:  Enables only IP applications  Can not support asymmetric traffic Backward notification  How does one standardize this mechanism for Bridges? Layer violations can make maintenance difficult (Support future changes in Upper Layers (IPv4, IPv6 etc.) Security challenges?

Page 17 Intel Corp. L2 Marking Mechanism proposal : L2-CI L2-CI (Congestion Indication)  Marking by bridges in L2 header during congestion Pros:  Standardized congestion notification mechanism in L2 networks  Clean layering, ULP-agnostic  L2-CI and TCP-ECN together provide hierarchical mechanism Equivalent to 802.1p and DSCP for CoS Cons:  Requires L2 header modification/extension for data frames  Requires End Stations to copy L2-CI information to ULP E.g. to IP-CE code-point for TCP flows to benefit

Page 18 Intel Corp. L2-CI: details

Page 19 Intel Corp. Layered view of network OS/ULP NIC Driver (MAC Client) 802 MAC Apps OS/ULP NIC Driver (MAC Client) 802 MAC Apps End Station Bridge (MAC Client) 802 MAC Switch 802 link

Page 20 Intel Corp. L2-CI : What it is and is not Is:  Mechanism for MAC Clients to provide congestion information  Enables MAC Clients to pass this information to upper layers (in end-systems typically) – API enhancements Enables triggering Rate Controllers in upper layers Is Not:  Does not define congestion detection mechanism for MAC Clients  Does not define Rate Controllers in MAC Client How to achieve:  Use CFI bit in Tag Header DE for Provider Bridge applications, CI for short-range networks  Definition of new L2 header (FESG can be leveraged)

Page 21 Intel Corp. DE and CI bit considerations Both mechanisms impact packets that “exceed traffic policy” DE: Packet is marked down making it eligible for drop in downstream switches  Primary target: Provider Bridge networks CI: Packet is marked so that sources can reduce injection rate  Primary target: Short range networks

Page 22 Intel Corp. AQM to detect congestion When AQM threshold is exceeded, mark the packets (e.g. with probability for RED) on L2 header to indicate that “this” packet experienced congestion  Actual position/s in header TBD Bridge Role:

Page 23 Intel Corp. Copy L2-CI information from L2 header Pass it to Upper Layer through API (enhanced)  E.g. NDIS API may need to be enhanced to carry additional information  Should be easier to handle in Chimney architecture for offload engines ULP = TCP/IP  IP to copy L2-CI information received via enhanced-API to IP-CE bit before handing to TCP flow  TCP remains unchanged (Sends ECN-response back etc.) ULP != TCP/IP  Use L2-CI information to propagate backwards towards the source Source can take appropriate Rate Controlling decisions End Node – MAC Client could also generate L2-CI End - Station Role:

Page 24 Intel Corp. L2-CI Considerations More than 1 bit congestion information  Congestion levels in the path (e.g. XCP)  Hook for reverse congestion notification (to be used by non-TCP protocols?) Additional information about “capabilities” of flow  Equivalent to “ECT” bit in IP – ECN  At congested devices, “non-capable” flows get packets dropped instead of marked

Page 25 Intel Corp. Summary In order to enable accelerated deployment of Ethernet into emerging limited-topology applications (clustering, backplanes, storage, data centers, etc.), IEEE should specify a standard mechanism for MAC Clients to provide congestion information to L2 edge devices, using wadekar_1_0501.pdf as a basis Any congestion notification mechanism defined by IEEE should be agnostic to L3-protocols  IP-CE is not agnostic to L3 protocols L2-CI mechanism provides ULP agnostic Congestion Notification for short range LAN topologies Modeling data for L2-CI with TCP-ECN shows that L2-CI can provide significant improvement in throughput and latency reduction for short-range networks Ref: