strange network hangs using kdb

strange network hangs using kdb

[John Jasen]

We have two similar systems, using acenic gigabit cards, plugged into a
3com Superstack 3 10/100/1000 switch.

When we use kdb on one of the systems, the other system stops receiving
packets.

We've seen this from kernel 2.4.2 with the kdb patches from oss.sgi.com to
2.4.4 with the patches.

Are there any thoughts on this? Any other information that may prove to be
useful?

--
-- John E. Jasen (jjasen1@umbc.edu)
-- In theory, theory and practise are the same. In practise, they aren't.

Re: strange network hangs using kdb

[Keith Owens]

On Tue, 5 Jun 2001 11:20:26 -0400, 
John Jasen <jjasen1@umbc.edu> wrote:
>When we use kdb on one of the systems, the other system stops receiving
>packets.

man linux/Documentation/kdb/kdb.mm, section Interrupts and KDB.

Re: strange network hangs using kdb

[John Jasen]

On Wed, 6 Jun 2001, Keith Owens wrote:

> On Tue, 5 Jun 2001 11:20:26 -0400,
> John Jasen <jjasen1@umbc.edu> wrote:
> >When we use kdb on one of the systems, the other system stops receiving
> >packets.
>
> man linux/Documentation/kdb/kdb.mm, section Interrupts and KDB.

I would expect one system to fall off the network, when it is put into
kdb. However, why does it have any effect on the other system, which may
or may not be in kdb?

Let's say we have two test servers (A and B); and four test clients (1-4).
1 and 2 are being used to test performance on A, while 3 and 4 are being
inflicted on B.

kdb is activated on A, and from that moment, no network traffic passes
between B and its clients, 3 and 4.

--
-- John E. Jasen (jjasen1@umbc.edu)
-- In theory, theory and practise are the same. In practise, they aren't.

Re: strange network hangs using kdb

[Keith Owens]

On Tue, 5 Jun 2001 12:20:25 -0400, 
John Jasen <jjasen1@umbc.edu> wrote:
>On Wed, 6 Jun 2001, Keith Owens wrote:
>> On Tue, 5 Jun 2001 11:20:26 -0400,
>> John Jasen <jjasen1@umbc.edu> wrote:
>> >When we use kdb on one of the systems, the other system stops receiving
>> >packets.
>>
>> man linux/Documentation/kdb/kdb.mm, section Interrupts and KDB.
>
>I would expect one system to fall off the network, when it is put into
>kdb. However, why does it have any effect on the other system, which may
>or may not be in kdb?

It does not.  kdb is intrusive when it trips but even it cannot reach
across a network and stop another machine.  Look for a networking
problem on the other system.

Re: strange network hangs using kdb

[slurn]

> 
> On Tue, 5 Jun 2001 12:20:25 -0400, 
> John Jasen <jjasen1@umbc.edu> wrote:
> >On Wed, 6 Jun 2001, Keith Owens wrote:
> >> On Tue, 5 Jun 2001 11:20:26 -0400,
> >> John Jasen <jjasen1@umbc.edu> wrote:
> >> >When we use kdb on one of the systems, the other system stops receiving
> >> >packets.
> >>
> >> man linux/Documentation/kdb/kdb.mm, section Interrupts and KDB.
> >
> >I would expect one system to fall off the network, when it is put into
> >kdb. However, why does it have any effect on the other system, which may
> >or may not be in kdb?
> 
> It does not.  kdb is intrusive when it trips but even it cannot reach
> across a network and stop another machine.  Look for a networking
> problem on the other system.
> 

Might the machine running kdb also be acting as a gateway or router
for the other boxen?  This would account for the lack of connectivity.

scott

Re: strange network hangs using kdb

[John Jasen]

On Tue, 5 Jun 2001 slurn@verisign.com wrote:

> Might the machine running kdb also be acting as a gateway or router
> for the other boxen?  This would account for the lack of connectivity.

Nope. No such luck. I _wish_ it was that easy. :)

As for it being a network problem, it is -- and completely reproducable.
Fire up kdb, and no-one receives packets. Otherwise, no problems occur --
even when both systems are under heavy network load.

--
-- John E. Jasen (jjasen1@umbc.edu)
-- In theory, theory and practise are the same. In practise, they aren't.

Re: strange network hangs using kdb

[Srikant Sharma]

Are you using eepro100 network cards ?
Because we have observed that many times when we enter kdb
the card on this machine goes crazy and starts pumping 8808 congestion
control packets and other cards stop sending data over the network.
try tcpdump and see if something like this is going on ..
--
Srikant



On Tue, 5 Jun 2001, John Jasen wrote:

> On Tue, 5 Jun 2001 slurn@verisign.com wrote:
>
> > Might the machine running kdb also be acting as a gateway or router
> > for the other boxen?  This would account for the lack of connectivity.
>
> Nope. No such luck. I _wish_ it was that easy. :)
>
> As for it being a network problem, it is -- and completely reproducable.
> Fire up kdb, and no-one receives packets. Otherwise, no problems occur --
> even when both systems are under heavy network load.
>
> --
> -- John E. Jasen (jjasen1@umbc.edu)
> -- In theory, theory and practise are the same. In practise, they aren't.
>
>