The multicast bug does appear correct, but the transfer of multicast data to
the other vsw's is very troubling

thanks,
 Greer


On 6/26/08, Raghuram Kothakota <Raghuram.Kothakota at sun.com> wrote:
>
> The following seems to be the issue that you encountered when
> the service domain crashed and rebooted.
>
> 6604983 multicast processing after a channel-reset is broken in vnet
>
> Greer Reichow wrote:
>
>> All,
>>  I'm tracing an interesting VSW issue that I'm hoping someone can help.
>>
>> EQ:
>> Router: Cisco 7206
>> Switch: Cisco 3750 - IGMP snooping turned off (all ports see all
>> multicast)
>> Server: Sun T5120 - 4 LDOMS - all OS'es solaris 08/07 patched to ldoms
>> 1.0.2
>> VSW Config: Host LDOM uses VSW0. Guest LDOMS 1 & 3 use VSW0 and VSW3 in an
>> IPMP active/passive pair. Guest LDOMS 2 & 4 use VSW1 and VSW2 in same style
>> IPMP pairing.
>>
>> We are a multicast driven environment, every application uses multicast
>> extensively for data sharing and information passing.
>>
>> Yesterday, the host LDOM crashed (cause under investigation) and didn't
>> reboot.  Once we booted the host, the VSW's were restored and unicast IP
>> info flowed again to the guest LDOMs.  However, no multicast was passing
>> from the switch down to guest LDOMs.  I tried to snoop VSW0 on the host OS,
>> but that failed (when is this going to be fixed?).
>>
>> Thinking on the issue, it occured to me that the guest OS wouldn't be
>> sending the IGMP join to the virtual switch as it had already sent it once
>> and believed the join  to still be active (not timed out).  However, since
>> the host crashed, it didn't have the connection request anymore.  I rebooted
>> guest ldom #1 and immediately multicast started flowing to ldoms 1 & 3 as
>> expected (since IGMP join would open the port on vsw0, and both guest OS'es
>> had already requested the multicast group).  However, when I checked ldoms
>> 2&4, they too were receiving the same multicast data!
>>
>>
>>
> I can't think of any reason why the ldoms2 &4 would receive the same
> multicast traffic, at least this doesn't look like a known issue.
>
> -Raghuram.
>
>> While in my standard environment, ldoms 2 & 4 should receive the data,
>> when the host crashed, the IGMP join request on the host was removed.  So
>> why would 2&4 receive the data when they too did not have an active join,
>> unless VSW's are passing data too each other when they shouldn't.
>>
>> Can anyone out there tell me:
>> 1)How VSW's communicate at the host level
>> 2)If the passing of multicast between them is expected
>> 3)How the IGMP joins/requests work in the VSW's?  Is the behavior similar
>> to Cisco's stack or a different implementation
>> 4)What tools are available to look at the VSW stack, or igmp sessions at
>> the VSW layer if we only have VSW0 defined at the guest OS?
>>
>> Thanks,
>>  Greer
>> --
>> This message was posted from opensolaris.org
>> _______________________________________________
>> ldoms-discuss mailing list
>> ldoms-discuss at opensolaris.org
>> http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss
>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
<http://mail.opensolaris.org/pipermail/ldoms-discuss/attachments/20080626/e33f39fc/attachment.html>

Reply via email to