The following data is a compilation of tcpdumps taken with
different client/server combinations. Each testrun generates three
data sets: HTTP/1.0, HTTP/1.1, and HTTP pipelining.
Naming Scheme
The names of the runs are named as follows:
<SERVER> - <BANDWIDTH> - <LATENCY> - <HTTP CLIENT VERSION>
- SERVER
- J stands for jigsaw and A stands for Apache. The
number is for us to keep track of which version of each we are
trying.
- BANDWIDTH
- H stands for high (for example a LAN) and L
stands for low, for example PPP
- LATENCY
- H stands for high latency, for example WAN) and L
stands for low, for example LAN
- HTTP CLIENT VERSION
- These are the suffixes used: http-10 is plain HTTP/1.0,
http-11 is without pipelining, http-11-pl is with
pipelining, and http-11-pl-zip is HTTP/1.1 with pipelining
and deflate content encoding.
Each combination is run 5 times and each run consists of a first
time retrieval followed by a cache validation. That is in all 10
times pr combination. We use this for naming the results:
- ft (first time - grabbing documents and headers)
- cv (cash validation - just getting the
headers)
As an example, take the following name:
A1-LH-http-11
- The server is Apache
- The bandwidth is low
- The latency is high
- The requests are non-pipelined HTTP/1.1
Data Summary
The data is furthur summarized at the end of
this file.
average: 370.0 packets 61887 bytes took
2.636 - overhead: 19.3%
average: 559.6 packets 248655 bytes
took 4.086 - overhead: 8.3%
average: 104.2 packets 14254 bytes took
4.434 - overhead: 22.6%