| Syren Baran 2007-05-31, 1:22 am |
| toby schrieb:
> On May 30, 4:36 pm, Syren Baran <s...@gmx.de> wrote:
Probably every server spawning a new Thread for every new connection,
instead of using a fixed number of worker threads. You could try WolfServer.
On the other hand, did any of of these _slow_ requests 2-90000 complete?
Given an average header size of 200 bytes this equates to 2000 seconds,
or more than half an hour. The graph seem to indicate one test for every
+500 connections, so over 160 tests for the graph or more than 320000
seconds or more than 88 hours.
Even your benchmark site says:[vbcol=seagreen]
>Interpretation
>
>Our figure shows the performance of a server when subject to parallel
>load. This kind of load is often generated in a so-called "Distributed
>denial of service attack".
>
>Apache dies at about 4,000 parallel sessions. Yaws is still
functioning >at over 80,000 parallel connections.
So its only save to say it can serve one connection during a DOS attack.
[deduction snipped]
>
> I don't think that deduction speaks to the question of concurrent
> scalability at all.
Right. It just points out a design flaw.
>
> This is an architecture problem. Of course C can be used to build much
> more scalable systems than Apache. My point, lost in the flames, was
> that Erlang/OTP happens to be one of them.
Apache uses a fixed number of threads. Edit source, change one number, e
voila.
This whole discussion is pretty theoretical anyway, do you know of any
life web server actually serving 90000 requests/sec? It must be
connected via a really fat pipe.
|