[OpenIndiana-discuss] weird packet garbling problem
Jim Klimov
jimklimov at cos.ru
Sat Feb 2 18:10:38 UTC 2013
On 2013-02-02 18:26, Jake Young wrote:
> Thanks Jim!
>
> I've seen similar behavior to what Ed described on a Solaris 10 Sparc box
> (V445) that is in a remote office.
I think we had something similar too on another setup, but the failure
involved larger packets, i.e. in SCP sessions timing out at 192KB or so.
Note the multiple of power of two in the problem ;)
I recall that the solution had to do with either MTU or the TCP window,
something was mismatched with the deployment's provider. I think it was
a smaller MTU on an intermediate link, and the overflow errors added up.
Maybe the firewall refused fragmented packets or some of the service
packets needed to dynamically reduce MTU, also...
IIRC, they ultimately forced a smallish MTU on the internet-facing
ethernet NIC (further connected to a DSL-Ethernet modem), and this
solved the problem for them...
Today I'd look at SSH-HPC extensions which may improve buffering and
windowing and such for the bulky SSH sessions. Maybe it can help the
running job adapt to changing networking conditions.
HTH,
//Jim
> I'll try that setting and see if it resolves my issue.
>
> I was VNCing in to an OI box in my office from my laptop via VPN and sshing
> in to the V445 box to kick off some backup scripts (which take a long time
> to run, due to the slow link to this office).
>
> Oddly enough I never had any issues with the automounter or nfs share I was
> backing up the V445 to. One issue was with the ssh connection between my
> OI desktop and the V445 dropping out (note there is very little text output
> in my backup script) after an hour or two. My other issue was having the
> VNC drop out between my laptop and desktop. It was setup as a persistent
> session, so no big deal, just annoying.
>
> I think the V445 interfaces are bge and the OI desktop is e1000g.
>
> Jake
More information about the OpenIndiana-discuss
mailing list