httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Aaron Bannert <>
Subject Re: mod_ssl as DSO
Date Tue, 06 Nov 2001 23:40:35 GMT
On Mon, Nov 05, 2001 at 09:31:29PM -0500, MATHIHALLI,MADHUSUDAN (HP-Cupertino,ex1) wrote:
> I don't think that's the actual problem.. The response varies from 38 tps
> (concurrency = 5) to 0.5 tps (concurrenty = 10) :-!.. I used to get 30 tps
> consistently on my box earlier (even with concurrency = 150).. Do you also
> see a similar degradation ?.. If no, there's some problem in my compiler /
> compiler options.. 

I am seeing this. Here's my test setup:

server: solaris 8 intel, worker MPM, statically built mod_ssl/openssl
client: linux 2.4, flood w/ssl hitting /index.html.en

Keepalive ON:     Concurrency    | Avg. Requests/second
                        2        |   7.39
                       10        |   6.88
Keepalive OFF:  -----------------------------------------
                        2        |   5.62
                       10        |   6.89

The accuracy of this test is not very high, since my client machine
is completely maxed out at concurrency of 10, while the server was not
ever maxed out (by CPU, at least). One important thing to mention is
that I was seeing significant spinning on rwlock acquires. The peak was
around 20/second (measured by mpstat in 5second intervals).

I am also very surprised that the performance with keepalive ON is hardly
better than without. It is indeed interesting that with Keepalive ON
the avg. req/sec actually *drops* when concurrency is increased. I
will definately be investigating this further.

> >> I get unresolved symbols error during the LoadModule phase. The reason I
> >> introduced this is because when mod_ssl is being loaded, it should bring
> in
> >> all the dependent libraries (libssl and libcrypto).. It also means that
> we
> >> shouldn't be linking -lssl, -lcrypto with httpd or any other executables.

I just thought I'd point out that I also get symbol errors at the LoadModule
phase. Mine happen to be relocation errors, but they are caused by the same
reason you pointed out. I'm having other issues with my compiler and linker
not playing nice, but if that turns out ok I will most likely apply your


View raw message