Draft Outline for "User Guide"

Editor: AnnHarding

Introduction

Goals and target readership

This is a snapshot of a "living document"

Related work

BestPracticeGuide

Performance Basics

User-perceived performance

Network performance metrics

The NetworkPerformanceMetrics topic includes detailed sub-topics on individual metrics that could be included here. Some of these sub-topics need refinement, others, e.g. OneWayDelay, are presented in quite some detail. I don't consider the detailed ones as a problem, as long as users can still find what they are looking for (skipping subsections is allowed).

Systemic view of performance problems

Bandwidth, delay, throughput

"Chattiness"

First steps at investigating performance problems

Problem isolation strategies

Measurement tools

Common errors

TCP Performance Primer

TransmissionControlProtocol

WindowBasedTransmission

FlowControl: slow-start, congestion avoidance etc.

EndSystemTuning, LargeTcpWindows

TCP Performance enhancements

Window scaling, timestamps

SelectiveAcknowledgements (SACK)

ExplicitCongestionNotification (ECN)

High-performance TCP variants

HS-TCP, H-TCP, BIC, FAST etc.

Hardware considerations

CPUs

Buses and system interconnects

Network adapters

File systems and disks

Examples

Compact benchmarking host with 1 Gb/s memory-to-memory TCP throughput

High-performance file/Web server for large audiences

Internet2 "Land-Speed Record" (2004) hosts

Operating system considerations

Out-of-the box system settings and tuning

Operating-specific tuning tips and tools

Application and protocol design

Bandwidth- and delay-tolerant protocols

Performance-friendly I/O interfaces

Avoiding blocking

Conclusions and further work

-- SimonLeinen - 20 Apr 2005