[e-nsp] x650 and MAC addresses missing in FDB ????

[Erik Bais]

Do you have specific type of filtering in place that puts the mac learning from hw on that switch  to sw based learning ? Like port security or someting else ?

That would explain the slower mac learning, the traffic flooding  etc. Eapecially in a large fdb environment.

Erik

You say it is originating from

Verstuurd vanaf mijn iPad

Op Dec 28, 2011 om 21:33 heeft "Marcin Kuczera" <marcin at leon.pl> het volgende geschreven:

> Erik Bais wrote:
>> Hi Marcin,
>>
>> Have you checked what kind of traffic it might be ?
>> Did you enable IGMP snoop and MLD snoop
>
> Both enabled
>
>> Is the traffic limited to 1 single vlan or is it a specific vlan / type traffic that causes this ? Microsoft NLB comes to mind ...
>
> It is not specific vlan, it is random behaviour.
>
> traffic is regular IPv4:
> 21:09:25.276836 00:30:88:19:32:a5 > 98:fc:11:cb:7b:ce, ethertype 802.1Q
> (0x8100), length 1518: vlan 3408, p 0, ethertype IPv4, (tos 0x20, ttl
> 53, id 49068, offset 0, flags [DF], proto TCP (6), length 1500)
>     216.155.151.93.80 > 188.137.52.217.49460: Flags [.], seq
> 322060:323520, ack 1, win 29, length 1460
> 21:09:25.276839 00:30:88:19:32:a5 > 98:fc:11:cb:7b:ce, ethertype 802.1Q
> (0x8100), length 1518: vlan 3408, p 0, ethertype IPv4, (tos 0x20, ttl
> 53, id 49069, offset 0, flags [DF], proto TCP (6), length 1500)
>     216.155.151.93.80 > 188.137.52.217.49460: Flags [.], seq
> 323520:324980, ack 1, win 29, length 1460
> 21:09:25.276841 00:30:88:19:32:a5 > 98:fc:11:cb:7b:ce, ethertype 802.1Q
> (0x8100), length 1518: vlan 3408, p 0, ethertype IPv4, (tos 0x20, ttl
> 53, id 49070, offset 0, flags [DF], proto TCP (6), length 1500)
>     216.155.151.93.80 > 188.137.52.217.49460: Flags [.], seq
> 324980:326440, ack 1, win 29, length 1460
>
> i.e. 98:fc:11:cb:7b:ce is not in FDB and after several seconds it
> appears, u-unicast traffic is gone for that host, but then I have
> another hosts with such traffic with no entry in FDB.
> Ok, in many cases it would be correct but not in ours..
>
> When particular DST MAC is broadcasted (unknown-unicast) it apears in
> FDB of next switch in chain towards DST (x450), then also I could see
> that age timer refreshes ! and - this MAC is still not in FDB of x650.
>
>
> On x650 I have almost 500 VLANs, 13k MAC addresses in FDB.
>
> I don't see this behaviour on x450 switches...
>
> Regards,
> Marcin
>
>
>
>
>
>
>
>> Regards,
>> Erik
>>
>>
>> -----Original Message-----
>> From: extreme-nsp-bounces at puck.nether.net [mailto:extreme-nsp-bounces at puck.nether.net] On Behalf Of Marcin Kuczera
>> Sent: woensdag 28 december 2011 20:39
>> To: extreme-nsp at puck.nether.net
>> Subject: Re: [e-nsp] x650 and MAC addresses missing in FDB ????
>>
>> Well,
>> I'am recalling my issue, maybe someone has a proper contact to someone from Extreme R&D.
>>
>> at the moment we have about 13k in FDB, and up to 25Mbit/s of unknown-unicasts traffic...
>> This is getting nasty...
>>
>> I need to know if x650 has a probablity that not all MAC may be represented in FDB (still below 32k limit).
>>
>> Regards,
>> Marcin
>>
>>
>>
>>
>> Marcin Kuczera wrote:
>>> Marcin Kuczera wrote:
>>>> hello,
>>>>
>>>> I have noticed an issue, that sometimes some mac addresses are not
>>>> visible in FDB on x650 stack.
>>>>
>>>> I have several x450 in chain-patch from customer and then x650 where
>>>> router is connected to.
>>>>
>>>> I could see MAC addresses for particular hosts in all x450 chain, but
>>>> it was missing in x650 for most of the time. Then sometimes it appeard.
>>>>
>>>> How do I catch them ? I'am monitoring unicast flooding on all vlans
>>>> and I have a lot of such cases...
>>>>
>>>>
>>>> EXOS release is 12.5.4.5 v1254b5-patch1-4 numer of MAC addresses in
>>>> FDB osscilates around 14k max, but this is not a problem because fdb
>>>> capacity in x650 is 32k.
>>>>
>>>> Nothing in logs..
>>>>
>>>> It is not tooo nasty, but I'am wondering if it is not some bug that
>>>> may affect some other services, i.e. multicasts.
>>> well, I spoke to an extreme engineer and he mentioned, that this might
>>> be some issue related to fdb hashing mechanism..
>>>
>>> It is supposed to be multi-level hashing, however it might be, that
>>> the lowest level "container" capacity, that contains whole MAC address
>>> - is limited.
>>> So, if there are some MAC addresses that will match this lowest level
>>> "container", and let's say - the limit per "container" is 16, and we
>>> have 17th MAC matching - it will not be placed there due to no space,
>>> and the traffic will be flooded...
>>>
>>>
>>> Is there anyone here that could verify this thesis ?
>>> How do I debug it ?
>>>
>>>
>>> Regards,
>>> Marcin
>>> _______________________________________________
>>> extreme-nsp mailing list
>>> extreme-nsp at puck.nether.net
>>> https://puck.nether.net/mailman/listinfo/extreme-nsp
>>>
>>
>> _______________________________________________
>> extreme-nsp mailing list
>> extreme-nsp at puck.nether.net
>> https://puck.nether.net/mailman/listinfo/extreme-nsp
>>
>