On Die 08.08.2006 21:16, A.D.F. wrote:
Alexander Lazic wrote:
As "A.D.F." wrote i agree with him/her?!
Without any doubt: "him" (the vi man).
Ah ok ;-)
What i also would prefer is to have a full descrition of the test
environment and setup ;-)
Yes, this is the core of the matter and of course everyone realizes
that these kind of tests are somewhat time consuming.
Sounds good that *all* know that such a benchmark couldn't be a quick
shot ;-)
Some issues which should be discussed from my point of view:
1.) is this list the right one to discuss this issue, i think we
shouldn't missued open && free resources, what mean the
list-owner about this issue?
Well, there are not many other lists about Cherokee.
Yep but as soon as we test some other server there could be ot of this
list, I don't know.
For me it's ok ;-)
2.) There are many java/perl/php/ruby/$HELL_LANGUAGE applications out
there, whould it be possible to make a general recommendation
which setting are the *best* for fcgi/scgi/http-proxy for $SERVER
and $APPL, yes i know this is a *big* target but why not try to
reach the stars ;-))
3.) What are some typical use-cases for
$APPLICATION(-Sever/Framwork/...)?
4.) What is the impact when you write access-logs or not and thru
which mechanism, if the $SERVER supported more then one
mechanism.
Yes, access-logs impact the performances but not as much as you can
think of (specially if log files and served files are located on
different disk devices); if the log handler is well optimized you have
at most a decrease of 5% - 15%.
Yep but how about '|' or 'fifos'?
It would be nice to say for example:
With accesslog you can get 1000 r/s without 1500 r/s.
With accesslog to '|' you can get 1200 r/s without 1500 r/s.
With accesslog to 'fifo' you can get 1200 r/s without 1500 r/s.
You got the point?
5.) Who *optimize* the underlaying OS after the first-run with many
steps?
I think that the only parameters that should be tweaked are the disk
and network ones; anyway you know that the best values depend on HW
capabilities, on OS and kernel version, etc.
Yep.
It would be also nice to can say:
With NFS you loose 10-15% or performance or not.
With proxying you loose 10-15% or performance or not.
And so on ;-)
6.) Which OS would be tested, because cherokee, lighttpd and nginx
support more then one OS with there preferd eventmodel?
Linux, FreeBSD, Mac OS-X, Sun Solaris (9, 10) and other Unixes;
Windows 2003 or Vista might be targets only in a far future.
Full ack ;-)
7.) Which tools are available on all OSes to measure CPU/Memory usage
and/or how comparable are this tools?
free(1), top(1) and similar tools.
Such as vmstat, mpstat, iostat, ... as i know free is not available on
all oses, isn't it?!
Thoughts, opinions?!
Yes, in the case of Cherokee don't expect too much from benchmarks
right now, I guess it will need at least another 8-12 (24 if I had to
be fair) months of development to gain optimal performances.
I have another point of view, because the *old* benchmarks speak another
word ;-)
Well let's go further with the hard facts ;-)
1.) Which benchmark tool(s)? should we use?
2.) How should be the 'client' and the 'server' should be connected?
3.) What is a Run and after how many steps is a run complete?
4.) Which tests are in how many steps per run?
5.) ...
6.) ...
regards
Alex
_______________________________________________
Cherokee mailing list
[email protected]
http://www.0x50.org/cgi-bin/mailman/listinfo/cherokee
