Alexander Gladysh wrote:
>> After looking at the C code for ab, it is not multi-threaded, true.
But it
definitely supports concurrent connections, using an event/poll model (just
like lighttpd). So it will certainly suffice for such testing.
I was worried by this article:
http://weblogs.java.net/blog/sdo/archive/2007/03/ab_considered_h.html
If I had more then one core on the client or server, this might concern
me too. Luckily I am unencumbered by such burdens.
But come on, 10 ms to send a GET request out a socket? Really? Now I may
run some high end P4s that are still about the fastest cores around, but
the time to "draining the socket of data, and sending the next request
on the socket" is measured in nanoseconds, not milliseconds. So in his
scenario the poll event model might add a total of about 1 - 3 ms to the
tenth request (if ab is badly written). Not 110ms.
Also his theory that "100 responses could come from one channel, or 1
response could come from each channel, and ab will give you the same
answer." is just plain incorrect. You can try it for yourself. If 91
(the max in this scenario, not 100) requests quickly came from 1 channel
and the other 9 took the whole test to complete, it would show 9
requests taking a very long time to complete, and 91 being quite fast.
Honestly, I don't think he has ever read the ab report past the rps
line... and certainly not looked at the code.
> I've got a feeling that you're trying (perhaps not conciously) to
> create your own ecosystem. Providing your own alternative
> implementations to existing modules sure does that. As the number of
> users (and, especially, the contributors) is limited, I feel that the
> software would have better chances to gain followers if it does not
> stray away from the mainstream too much.
I am trying to make a system to increase the transaction rate for my
employer without buying new servers and maintaining reliability.
Everything else is secondary. I am open-sourcing it so that anyone else
can use it if they want. We will never sell software, but this way if I
get hit by a bus someday there is a chance that someone else out there
is familiar with what the company is using.
Now if I can build it in such a way that makes it more useful for
others, but does not impact the usefulness to my benefactors, then I
will certainly try :)
_______________________________________________
Kepler-Project mailing list
[email protected]
http://lists.luaforge.net/cgi-bin/mailman/listinfo/kepler-project
http://www.keplerproject.org/
