John Doe wrote:
Talking about benchmarks; I did get around 770 r/s and was wondering why...
In fact, it was just because my test object was 15KB.
With a 1KB object, I get 5000+ r/s.
So, what is the "usual" size of test objects for the benchmarks...?
If you're running benchmarks then can you please provide some CPU and
IO profiling where appropriate?
I would like to know why a 15KB object cops out at 770req/sec but a
1KB object cops out at 5000 req/sec.
Should have added that my 100Mb switch must be the reason: 770*15*8 ~= 92400 Mb/s
My benchmarks are pretty basic for now (always the same object) so not really worth mentioning.
I might give some in the future when the setup is finalized.
I was just curious about the effect of objects size choice in others benchmarks...
The smaller, the faster... no?
Probably.
The only 'recent' results contributed so far have been spot checks taken
from real-world caches. So likely to have a wide range of objects. And a
likewise large error margin.
Amos
--
Please use Squid 2.7.STABLE4 or 3.0.STABLE8
