|
> Where can get any benchmarks to test LVS? Is there anyone
> do benchmark research in LVS group?
You can use LVS to benchmark anything really. If you set it up with Direct
Routing, you're almost certain to be bottlenecked at the real servers, not
at the IPVS director box.
You could create artificial benchmarks to try to test IPVS, but all the
industry standard benchmarks I know will task the servers, not the IPVS
director.
I'm running benchmarks now for my employer. We've run IPVS/DR on
clusters from 1 to 16 machines of webservers (LinuxWorld, if you saw the
VA cluster there), and the request traffic on
the director was still not significant, and the processor usage is usually
laughable.
Since http traffic is usually at least 10:0 response-to-request, you
should expect to be able to use an IPVS box pulling 100Mbit of requests
to forward to real servers returning 1GBit of responses. One IPVS box
*can* do this under benchmarks, and that amount of traffic makes you
a Yahoo. :)
In short, I've found it better to focus on the http/ftp benchmarks on the
real servers than the IPVS box.
Anyone else have experience with high-traffic IPVS benchmarks?
-drew
-------
Drew Streib <d@xxxxxxxxxxx> 408.542.5725
Sr Developer, Community Liason, SourceForge | <dtype@xxxxxxxxxxxxxxx>
System Administrator, Linux International | <dtype@xxxxxx>
Information Architect, VA Linux Systems | <dtype@xxxxxxxxxxx>
Admirer, Occasional Programmer, Linux.com | <dtype@xxxxxxxxx>
|
