[Linux] Network stalls

[linux@flux.org]

Quoting Peter Capazzi <pcapazzi@dwcsolutions.com>:

> Actually... same box.
>
> It's a blade server. 4 blades. 3 of them I own and have identical OS'
> installed (Suse 10).
>
> Blade1 copy to Blade2 -> 30-40 MB/s dipped occassionally to 18... but
> recovered well back into 30-40.
> Blade1 copy to Blade3 -> 30-33 MB/s dipped to 12-14 MB/s... doesn't
> recover... if anyting goes down again to 5-7... then to KB's.
>
> It's inconsistent... the time at which the bandwidth drops is different
> each time... sometimes it gets down to 12-14 and manages to stay there.
>

  Doh.. accidentally hit SEND..
Mike suggested checking the disk I/O.  This is usually what the =20
culprit it. The system will buffer transfers and then write it out. =20
The write-out can lead to pauses.  Have you tried doing a null =20
transfer?? Try ftp'ing or netcat'ing (not ssh since there is some =20
overhead) to the other machine with /dev/null as the destination.  If =20
you can send a few gigs without error then it's likely to be the disks.

You can also try using blowfish encryption with SSH if it turns out =20
not to be the disk. It's slightly less overhead on the CPU.