[DRBD-user] connect error -22 with SDP/InfiniBand

Lars Ellenberg lars.ellenberg at linbit.com
Fri Sep 17 22:11:46 CEST 2010


On Fri, Sep 17, 2010 at 02:12:35PM -0500, J. Ryan Earl wrote:
> >
> > If that gets you connected, then its that bug.
> > I think I even patched it in kernel once,
> > but don't find that right now,
> > and don't remember the SDP version either.
> > I think it was
> > drivers/infiniband/ulp/sdp/sdp_main.c:addr_resolve_remote()
> > missing an (... || ... = AF_INET_SDP)
> 
> 
> Is this the fix to which you refer?
> http://www.mail-archive.com/general@lists.openfabrics.org/msg10615.html

That's certainly relevant as well,
but it would have returned EAFNOSUPPORT, which is 97.


I doubt that this has anything to do with performance, btw,
it is just the address lookup during connect.

My guess is if you strace a netcat in userland, using your sdp preload
thingy, you'll likely see that it only creates the socket as
AF_INET_SDP, but all the rest of the network functions keep using
AF_INET, so no-one ever noticed. If that is intentional, we'd have to
adjust that in DRBD. If not, it needs to be fixed in the sdp stack.

DRBD over SDP performance tuning is a bit tricky,
and no, I don't remember the details, it's been a while.

I think cpu usage dropped considerable, thats a plus. But neither
single write latency nor sequential throughput of a single connection
improved much or even degraded respective to IPoIB.  If you have
multiple DRBD, thus multiple connections, the cumulative throughput
scaled better, though.

But please go ahead and tune your stack, on your hardware, which may be
more capable than the test lab hardware we used.
Depending on your hardware, and the quality recommendations of
your SDP tuning expert, your findings may be different.

-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.
__
please don't Cc me, but send to list   --   I'm subscribed


More information about the drbd-user mailing list