RE: [Xen-users] iscsi conn error: Xen related?

["Ross S. W. Walker" <rwalker@xxxxxxxxxxxxx>]

Fred Blaise wrote:
> 
> Hello all,
> 
> I got some severe iscsi connection loss on my dom0 (Gentoo 
> 2.6.20-xen-r6, xen 3.1.1). Happening several times a day.
> open-iscsi version is 2.0.865.12. Target iscsi is the open-e 
> DSS product.
> 
> Here is a snip of my messages log file:
> May  5 16:52:50 ying connection226:0: iscsi: detected conn error (1011)
> May  5 16:52:51 ying iscsid: connect failed (111)
> May  5 16:52:51 ying iscsid: Kernel reported iSCSI connection 226:0 error 
> (1011) state (3)
> May  5 16:52:53 ying connection215:0: iscsi: detected conn error (1011)
> May  5 16:52:53 ying iscsid: connect failed (111)
> May  5 16:52:53 ying iscsid: connect failed (111)
> May  5 16:52:53 ying iscsid: connect failed (111)
> May  5 16:52:53 ying iscsid: connect failed (111)
> [...]
> 
> and sometimes:
> May  5 16:53:11 ying iscsid: connection227:0 is operational after recovery (6 
> attempts)
> May  5 16:53:11 ying iscsid: connection221:0 is operational after recovery (6 
> attempts)
> May  5 16:53:12 ying iscsid: connection214:0 is operational after recovery (9 
> attempts)
> 
> Usually, this means loss of my Windows HVM machines.. paravirtualized 
> machines seem to handle that ok, oddly (qemu?).
> 
> I have read that this could be due to network state change/asymetric 
> routing.. but dunno really in my case. I have 4 network interfaces (2 
> dualport cards, Intel PRO/1000 MT):
> 
> - 1 is dedicated to storage, with jumbo frames enabled.
> - 1 for admin tasks (web interface, ssh)
> - 2 for various vlans used
> 
> Anyone experienced this already? Found a solution? Any 
> recommendations? 
> Any help much welcome.

Try disabling jumbo frames. I have seen a lot of cases of jumbo
frames causing a stall in the switch ports on some switches. Also
if using jumbo frames, make sure flow control isn't a problem as
a lot of switches have inadequate port buffers to handle flow
control and jumbo frames.

To note: Jumbo frames on 1Gbe isn't necessary and will in fact
increase latency which will decrease throughput. Jumbo frames
is really meant to reduce interrupts and is a lot more
affective with 10Gbe then 1Gbe. On 1Gbe if interrupts are
running too high I would use interrupt coalescence as a first
attempt to reduce them.

-Ross

______________________________________________________________________
This e-mail, and any attachments thereto, is intended only for use by
the addressee(s) named herein and may contain legally privileged
and/or confidential information. If you are not the intended recipient
of this e-mail, you are hereby notified that any dissemination,
distribution or copying of this e-mail, and any attachments thereto,
is strictly prohibited. If you have received this e-mail in error,
please immediately notify the sender and permanently delete the
original and any copy or printout thereof.


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users