[Spread-users] high cpu load

George Schlossnagle george at omniti.com
Fri Aug 31 10:07:50 EDT 2001


Hi Dirk,

This is almost certainly not a problem with mod_log_spread, it seems to
definitely be a problem in your spread rng itself.

Yair, is it possible to provide a patch whihc will do verbose logging about
who the retran's are going to?  It might make debugging the problem a
wee-bit easier.

George

----- Original Message -----
From: "Dirk Vleugels" <dvl at 2scale.net>
To: <spread-users at lists.spread.org>
Cc: "Dirk Vleugels" <dvl at 2scale.net>
Sent: Friday, August 31, 2001 6:05 AM
Subject: Re: [Spread-users] high cpu load


> Hello,
>
> [stripped the mail recipients]
>
> On Thu, Aug 30, 2001 at 10:20:42PM -0400, Yair Amir wrote:
> > > I have still not found an explanation. There is no network
> > > problem on a lower layer i think.
> >
> > There is definitely a problem with the too high retransmission.
> >
> > > > > sent pack: 5706975      recv pack : 17722855    retrans    :
5540933
> > > > > u retrans: 5386778      s retrans :  154155     b retrans  :
0
> >
> > >From this I understand that you probably have only one segment in you
file, which is fine.
> > u retrans is when only one guy lost it so far so you unicast to this
guy.
> > s retrans is when at least 2 guys from the same segment lost it (again,
it seems
> > you have only one segment in the system). BUT these numbers are
extremely high.
> > Something is really wrong. We have some programs you can find in the
distribution
> > called "s" and "r" that can show what happens in terms of low level
issues on
> > your network. s and r are not part of the Spread toolkit really but they
can be
> > useful to us when we try to see if there is a problem on the udp level.
> > Maybe you can poke a bit there.
>
> Using the same network spread is running on:
>
> Send:
>
> ./s -a 192.168.101.127 -t 2 -s 1000 -b 50 -n 100000
> [.....]
> total time is (20,466537), with 0 problems
>
> Receive:
> ./r
> Ready to receive on port 4444
> -------
> Report: total packets 100000, total missed 16, total corrupted 0
>
> It's no udp problem i guess. I will take a closer look on
> mod_log_spread.
>
> Cheers,
> Dirk
>
>
> _______________________________________________
> Spread-users mailing list
> Spread-users at lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users
>







More information about the Spread-users mailing list