performance

[Yussef Alkhamrichi <ya...@hotmail.com>]

Hi Guys,

I've no idea if anyone is using httpd CLI yet, I wanted to inform about its 
use. I think it is a great project combining the .NET Framework (and myabe 
mono) and Apache. I have developed in C# for the last 3 years, everything 
from windows apps, services and websites.

I've pulled a version from the subversion repository a couple of weeks ago 
and got it compiled and running with two minor changes.

Changed

if (!repl || wcscmp(repl, L"\\apache.exe"))

into

if (!repl || !wcscmp(repl, L"\\apache.exe"))

and changed the PublicKeyToken refrence on line 329 of mod/mod_aspdotnet.cpp 
to match the assembly.

I've stress tested it quickly, the strange thing was that the req/sec were 
lower than on IIS but the throughput in mb/sec was higher. I haven't looked 
any further, but to bad the number of requests is lower than with IIS. Maybe 
I missed a Apache configurational setting or this will be improved in 
feature releases.

Does anyone have experiences with the perfomance of this module ? (Maybe 
it's the wrong place to ask it on this mailing list, sorry for that)

_________________________________________________________________
On the road to retirement? Check out MSN Life Events for advice on how to 
get there! http://lifeevents.msn.com/category.aspx?cid=Retirement

Re: performance

["William A. Rowe, Jr." <wr...@rowe-clan.net>]

At 03:37 PM 10/7/2004, hammett wrote:
>From: "William A. Rowe, Jr." <wr...@rowe-clan.net>
>
>> There is one interesting consideration... do we want two separate
>> keys, one dev key (the one in there now) used by anyone who builds
>> this package themselves (unless, if they would like they can create
>> their own) ... and the other used for official binary release (and
>> held by the release manager alone)?
>
>This is a pain, but seems like a one viable strategy. A better strategy
>would be to have the key on the SVN but not publically available.

When we deal in pgp key files, we countersign one another's keys
but maintain strict possession of our own.

I'm thinking that if we have, instead of a KEYS file, another master
file containing the keys of all release managers.  Anyone can use the
public key for their own -dev builds, or stuff in their own, but
either way the mod_aspdotnet must match to the Apache.Web sk file.

The big pain would be if a user tried to build -only- mod_aspdotnet
or Apache.Web themselves.  At that point they would be out-of-sync.

Of course, with a bit of documentation in README this problem could
be dispensed with right away.

Re: performance

[hammett <ha...@uol.com.br>]

----- Original Message -----
From: "William A. Rowe, Jr." <wr...@rowe-clan.net>
To: "CLI Dev" <cl...@httpd.apache.org>
Sent: Thursday, October 07, 2004 8:30 AM
Subject: Re: performance


> There is one interesting consideration... do we want two separate
> keys, one dev key (the one in there now) used by anyone who builds
> this package themselves (unless, if they would like they can create
> their own) ... and the other used for official binary release (and
> held by the release manager alone)?

This is a pain, but seems like a one viable strategy. A better strategy
would be to have the key on the SVN but not publically available.

Cheers,
hammett

Re: performance

["William A. Rowe, Jr." <wr...@rowe-clan.net>]

At 07:59 AM 10/7/2004, Yussef Alkhamrichi wrote:
>Hi Guys,
>
>I've no idea if anyone is using httpd CLI yet, I wanted to inform about its use. I think it is a great project combining the .NET Framework (and myabe mono) and Apache. I have developed in C# for the last 3 years, everything from windows apps, services and websites.

Glad you are here, glad you gave it a spin :)

>I've pulled a version from the subversion repository a couple of weeks ago and got it compiled and running with two minor changes.
>
>Changed
>
>if (!repl || wcscmp(repl, L"\\apache.exe"))
>
>into
>
>if (!repl || !wcscmp(repl, L"\\apache.exe"))

I'm removing the later test - it was both a path check and a look 
for a proprietary flavor of the httpd.exe binary.  The test for 
(!repl) is sufficient, since we don't care if they name the Apache 
binary as httpd.exe, apache.exe or something else.

>and changed the PublicKeyToken refrence on line 329 of mod/mod_aspdotnet.cpp to match the assembly.

Also committed, thanks.

There is one interesting consideration... do we want two separate
keys, one dev key (the one in there now) used by anyone who builds
this package themselves (unless, if they would like they can create
their own) ... and the other used for official binary release (and
held by the release manager alone)?

>I've stress tested it quickly, the strange thing was that the req/sec were lower than on IIS but the throughput in mb/sec was higher. I haven't looked any further, but to bad the number of requests is lower than with IIS. Maybe I missed a Apache configurational setting or this will be improved in feature releases.

IIS uses fibers (smaller than threads) and can internally cache its
results.  (The mod_cache code -is- improving radically, I expect you
can get similar results with this environment.)  I'd be most interested
if you were handing off requests serially, or were testing load against
10 or 100 parallel clients.  ab.exe distributed with Apache allows you
to specific -c 10 for 10 concurrent requests, etc.

>Does anyone have experiences with the perfomance of this module ? (Maybe it's the wrong place to ask it on this mailing list, sorry for that)

Due to Microsoft's license terms of IIS, etc, Covalent never published 
or made claims to the performance of this module when it was developed.

Remember too - the first request will be slower while the assembly
module is first loaded, and compiled from IL to the native code cache.

Bill