Design, Implementation and Evaluation of a Client Characterization Driven Web Server

1
Design, Implementation and Evaluation of a Client
Characterization Driven Web Server

• Balachander Krishnamurthy ATT Research Labs
• Craig Wills WPI
• Yin Zhang ATT Research Labs
• Kashi Vishwanath Duke University

2
Motivation

• Users with varying connectivity request a variety
  of files
• Web site wants no user to get disinterested
• Handle requests based on connectivity

3
Outline of our solution

• Figure out connectivity use web server logs
• Classify clients(IPs) store in shared memory
• Cluster group of IPs KW00
• When a page is requested, the Web server
• Identifies client connectivity
• Takes action if required alternate content etc.

• 122.3.4.22 - - 12/Apr/2003014022 0400
  GET /foo.html HTTP/1.0 200 147

4
System Architecture
Server Log
Classifier
User Requests
Classification Table (Shared Memory)
Apache Processes
5
What is a page access?

• User clicks container page page eg. /foo.html
  
• Subsequent requests embedded objects
  eg. /img1.gif
• At server how long before the user
• starts getting the first object ?E_first
• gets the entire page ? E_last
• Smoothed means for popular pages
• Classes (poor, rich, normal) KW02
• Poor E_first gt3 and E_last gt5
• Rich E_first lt1 and E_last lt2

6
Modified Apache Server

• Modification to Apache 1.3.24
• three files, less than 200 lines
• http_main.c, http_protocol.c, http_core.c
• One time overhead
• Open cluster library
• Read in configuration file
• /foo.html .gz .p_lc .r_mc
• Initialize shared memory

7
Modified Apache Server Per Request Changes

• Per Request Changes
• Is it a popular page ( URI ) ?
• Lookup class (in shared memory data)
• Cluster lookup if IP unavailable
• Tailored action if appropriate and log it

8
Testing How to choose test pages

• Proxy logs from a large manufacturing company
  (Dec 2001)
• Over 100,000 users
• Select 1000 most popular pages
• Download these April 2002
• 641 successful pages reconstructed
• 33 and 66 percentile value for characteristics
• Container bytes, of Embedded object, embedded
  bytes
• 3 x 3 x 3 27 buckets of pages

9
TestingDistribution for test pages
Percentage of pages in each bucket
10
Server Overhead Latency increase at
client

• Generate IPs (X-IP header)

11
Stress Test

• Plot of average increase in processing time for
  the modified server

12
Placing Clients and Modified Servers

• Prototype Apache Server with our test site
• Linux att.com in NJ, USA
• Linux wpi.com MA, USA
• FreeBSD icir.org CA,USA
• Clients
• attATT Labs-Research, NJ,USA
• de Saarbruecken University, Germany
• cable cable modem user, NJ,USA
• modem 56Kbps dialup modem user, NJ,USA
• ukLondon, U.K via a dedicated 56Kbps line.

13
ClientsObserved network
characteristics
uk-icir
Spans a wide spectrum
uk-att
cable-icir
de-att
att-icir
att-wpi
uk-wpi
14
Experiments Httperf Clients

• Request similar mix of pages as described earlier
  
• 200 random requests each
• New Headers X-Server-Actions, X-Class
• Baseline measure
• parallel-1.0 up to 4 parallel HTTP/1.0 requests
• Server actions
• Manner of delivery
• Compress, serial-1.1, pipeline-1.1, bundle
• Amount served
• Base-only, half-object,half-resolution

15
Classifier Stability and Correctness
Rich
Moderate
Poor
Overall consistent with
expectation 90
16
Results Low throughput client

• Uk-wpi

17
Results medium throughput

• de-att

18
ResultsSummary
19
Conclusions

• What all do we do ?
• Online client classification
• Deliver modified server actions
• Measure latency reduction to different clients
• Compare various actions
• First to do this in a unified framework.

20
Conclusions Server and Classifier

• Overhead at server
• Average75 usec. Negligible for end-user
• Turn off classification during overload
• Server poor-content during overload
• Classification
• Close to expectation
• Stable over the duration of experiment
• Improve by using select pages and better
  thresholds

21
Future

• Other server actions
• Delta encoding
• Policies regarding cacheablility of objects
• Create a test for clustering

22
Acknowledgement

• Client testing
• Saarbruecken, Germany
• ICIR, USA
• Proxy logs - Manufacturing company

23
Thanks!
Slides http//www.cs.duke.edu/kvv/www.ppt

• Questions ?

24
Related Work

• Mark network packets NT02
• Improve performance of all clients
• Adapt content based on server load AB99
• Users expectation BBK00
• Alternate admission control and server scheduling
  policies CMC01, CR02
• Dynamically altering multimedia content MB01
• Explicit client notification e.g. SPAND SSK97