performance benchmarking

[rajks]

Hi,

I am using on linux with 8 cores
   1. The Http file server example

   2. "httperf" benchmarking tool from HP labs - 
http://www.hpl.hp.com/research/linux/httperf/

       I had to temporarily change the FD_SETSIZE to higher number 65536 to
allow httperf to compile with higher fd set size, otherwise the default is
1024.
       /usr/include/bits/typesizes.h:#define __FD_SETSIZE  65536

   3. "autobench" perl wrapper for httperf  -
http://www.xenoclast.org/autobench/ 

       Autobench runs httperf a number of times against each host,
increasing the number of requested connections per second on each iteration,
and extracts the significant data from the httperf output, delivering a CSV
or TSV format file which can be imported directly into a spreadsheet for
analysis/graphing.


I ran my tests with autobench with start rate of 10 r/s to 100 r/s at step
rate of 10 r/s for each test at 120 secs period on a static file of 500k
size.  The req rate, conn rate and reply rate typically should be same or
close up to some threshold. But here at request rate 20 r/s the stats are
deviating. When I try the same with say the NGINX/Lighttpd on same static
files the stats are consistent upto 60 r/s before the saturation.

Also with netty when lots of connections are transferring data, every new
connection from another machine takes a lot of time and transfer time of
body also takes a lot of time.

The transfer time of body reported by httperf also increases to a higher
number between 10 r/s and 20 r/s test.

Is there something I should be tuning in netty to have a consistent scaling
?

Has somebody used httperf to benchmark netty apps and how they fixed their
netty apps to work better ?

Thanks
Raj

-- 
View this message in context: http://n2.nabble.com/performance-benchmarking-tp3971150p3971150.html
Sent from the Netty User Group mailing list archive at Nabble.com.