Performance Issues in WWW Servers - PowerPoint PPT Presentation

1 / 14
About This Presentation
Title:

Performance Issues in WWW Servers

Description:

Optimize the frequency or cost of the blow operations: Network Computing Lab. 4 ... Macrobenchmark : load the server with concurrent requests for a range of files ... – PowerPoint PPT presentation

Number of Views:16
Avg rating:3.0/5.0
Slides: 15
Provided by: nclabK
Category:

less

Transcript and Presenter's Notes

Title: Performance Issues in WWW Servers


1
Performance Issues in WWW Servers
Erich Nahum, Tsipora Barzilai, and Dilip
Kandlur ACM SIGMETRICS Conference on Measurement
and Modeling of Computer Systems, May, 1999
  • 2001. 12. 4
  • Suh, Young-Ho

2
Introduction
  • Goal
  • Improve support for high-performance WWW servers
  • Solution approach
  • Optimize the operating system and the network
    protocol stack
  • New socket functions
  • acceptex(),transmitfile() / send_file()
  • Do they provide any performance benefit?
  • Does transmitfile/send_file() show any
    improvement over the already available mmap() and
    writev() system calls?
  • Per-byte optimizations
  • How well can we approximate a zero-copy
    integrated I/O architecture, while continuing to
    exploit the benefits of existing file systems?
  • What sort of performance impact will eliminating
    data touching operations have for WWW servers?
  • Per-connection optimization
  • How can the per-connection overhead be reduced,
    without violating the TCP protocol sepecification?

3
Background and Related Work(1/2)
  • Optimize the frequency or cost of the blow
    operations

4
Background and Related Work(2/2)
  • Optimize OS performance directly
  • Kaashoek et al.
  • Customized OS (HTTP server OS)
  • Druschel, Pai, and Zwaenopol
  • Extensible micro-kernel OS
  • Integrated I/O system for UNIX called I/O Lite
  • This paper
  • Like Druschel, Pai, and Zwaenopol, they believe a
    general-purpose OS is necessary to support a wide
    variety of services over HTTP, such as dynamic
    content
  • However, they try to minimize changes of the API

5
Experimental Setup and Testbed(1/2)
  • Hardware
  • 4 IBM 43P RS/6000
  • 128MB RAM, 200MHz PowerPC 604e processor
  • On-chip instruction and data caches, 32KB 4-way
    associativity
  • 1MB direct-mapped secondary cache
  • Workload Generator
  • WebStone
  • Microbenchmark load the server with many
    concurrent requests for the same file
  • SURGE
  • Macrobenchmark load the server with concurrent
    requests for a range of files

6
Experimental Setup and Testbed(2/2)
  • Operating System
  • AIX version 4.3.1
  • With the addition of several modifications(increme
    ntal)
  • acceptex(), send_file(), modified TCP protocol
    implementation
  • WWW Server
  • The fastest WWW Server Flash-Poll (incremental
    modification)

7
Results(1/7)
  • Evaluating the Proposed Socket Functions

8
Results(2/7)
  • Evaluating Per-Byte Optimization

9
Results(3/7)
  • Evaluating Per-Connection Optimizations

10
Results(4/7)
  • Evaluating Per-Connection Optimizations

11
Results(5/7)
  • Evaluating Per-Connection Optimizations

12
Results(6/7)
  • Evaluating Per-Connection Optimizations

13
Results(7/7)
  • Evaluating Overall Performance with SURGE

14
Conclusion
  • New socket functions
  • Per-byte optimizations
  • Per-connection optimizations
  • Aggregate benefits
Write a Comment
User Comments (0)
About PowerShow.com