Date: Thu, 18 Apr 1996 11:11:33 -0400
Message-Id: <199604181511.LAA01684@luers.qosina.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
To: new-httpd@hyperreal.com
From: Aram Mirzadeh <awm@qosina.com>
Subject: Re: socket failed
Sender: owner-new-httpd@apache.org
Precedence: bulk
Reply-To: new-httpd@hyperreal.com
Status: O

At 09:41 PM 4/16/96 MDT, you wrote:
> 
>> Ammm... I hate to bring this up now that the votes are due in, but it seems
>> with each release of apache we're increasing the socket error to hit ratio 
>> here. 
>> 
>> >From 11am on the 15 to 10pm on the 16, I generated over 132 socket errors
>> thi sis for about 3132 hits.  That's 30:1 ratio... with 1.0 it used to be
>> about 50:1.   
>
>Out of curiosity, do things get better or worse if you disable the
>accept serialization code?

I tried it both ways...  w/o part just about last night to this morning.... no 
major change... I had about 12 the night before for about 250 hits, and 
last night I have about 9 for 197.

>Didn't that Linus character say that having all processes be ready to
>accept() was a good thing?

It's done..  I'm just waiting for the 2.0 kernel to come out to upgrade.  But
as I hear from the people who are using the beta, apache still reporting
socket errors.  <shrug>

>The networks are hosed, so you might be seeing more 'there one minute
>gone the next clients'.

I agree.  The only problem I have is other server don't either report it, or 
aren't effected as much by it. 

>You could always upgrade to FreeBSD  ;-)

I have a FreeBSD server running.  It's our NFS server.   I have to say 
Linux's NFS server ( locking ) simply sux. 

<Aram>

--
Aram W. Mirzadeh, MIS Manager, Qosina Corporation
http://www.qosina.com/~awm/, awm@qosina.com
Apache httpd server team http://www.apache.org