2006-12-08 17:41:57 +00:00
|
|
|
Some ideas (contributions/patches welcome):
|
|
|
|
|
|
2007-05-29 05:36:13 +01:00
|
|
|
- port to libevent to improve scalability and deal with the file descriptor cap
|
|
|
|
|
- Add option to output results as rdf (xml), cvs or default
|
|
|
|
|
- Add ability to read entire POST and GET messages from logs and send them
|
2006-12-08 17:41:57 +00:00
|
|
|
- wsesspage: don't fetch same object more than once (assume the existence
|
|
|
|
|
of a cache)---this avoids trouble with recursive pages
|
|
|
|
|
- make httperf easier to use; some ideas:
|
|
|
|
|
o Make httperf into a network daemon that is controlled by
|
|
|
|
|
an httperf frontend. This would allow running tests with
|
|
|
|
|
a single command line even when multiple clients are involved.
|
|
|
|
|
The performance results should be reported both on a
|
|
|
|
|
per-client basis and in a summarized form reporting overall
|
|
|
|
|
server behavior.
|
|
|
|
|
o Provide (default) scripts to run certain benchmarks.
|
|
|
|
|
o Provide (default) scripts to produce performance graphs.
|
|
|
|
|
- use cycle registers to get time on CPUs that have such registers
|
|
|
|
|
(IA-64, PA-RISC, x86, Alpha, at least)
|
|
|
|
|
- randomize the uri's generated by --wset so we can defeat file prefetching
|
|
|
|
|
that the server (or the server's OS) may attempt to do
|
|
|
|
|
|
|
|
|
|
Done:
|
|
|
|
|
|
|
|
|
|
+ Specifying --session-cookie without specifying a session workload causes
|
|
|
|
|
httperf to core-dump (reported by Dick Carter, 10/13/98)
|
|
|
|
|
+ elevate `Session' to same level as Call and Connection
|
|
|
|
|
+ sample session throughput when using sessions (not just rate throughput)
|
|
|
|
|
+ integrate Dick's wsesslog
|
|
|
|
|
+ write man-page
|
|
|
|
|
+ chunked replies fail after the first reply (at least on HP-UX)
|
|
|
|
|
+ core.c does not fully support pipelined requests yet; would require
|
|
|
|
|
changes to the watchdog management
|
