[nsp] Routing through Management Vlan on 3750?

[Deepak Jain]

It turns out the ASIC is dropping the traffic on Queue 11 which I am 
told by TAC is for "unknown" traffic types. Caching makes the problem 
worse for some reason. It is still under active investigation.

Happy New Year.

Deepak

Jeff Nelson wrote:
> I appreciate you following up on your own post. I've been looking to use the 3750 for some one-off scenarios and with every new device comes some new "features". I'll file this one.
> 
> --jeff
> Deepak Jain(deepak at ai.net)@03/12/16 23:27:
> 
>>I know I'm replying to my post, but since these things seem to have a 
>>long shelf life in google....
>>
>>I opened a TAC ticket on this issue, and it looks like the 3750 might 
>>have some problem with caching on these interfaces. The packet loss and 
>>problem disappeared once caching was turned off. YMMV.
>>
>>DJ
>>
>>Deepak Jain wrote:
>>
>>
>>>On a 3750,
>>>
>>>g1/0/24 - g1/0/27 are L2 ports that bring traffic into the box over VLAN 1.
>>>
>>>g1/0/28 is the uplink (no switchport, ip addr x.x.x.x)
>>>
>>>There is a default route to the uplink on g1/0/28.
>>>
>>>int vlan 1
>>>has several ip addresses configured corresponding to all of the networks 
>>>on g1/0/24-g1/0/27.
>>>
>>>For numerous legacy reasons, the address allocations and port 
>>>allocations don't easily correspond to discrete subnet masks.
>>>
>>>No ip addresses are configured on loopback 0 or anywhere else.
>>>
>>>Pinging/tracing from the 3750 to the rest of the internet is fine.
>>>
>>>Pinging from the 3750 to any of the hosts on g1/0/24 - g1/0/27 seems 
>>>fine, at HIGH packet rates, some packet loss is noted -- could be the 
>>>server or it could be the config. But it leads to the question:
>>>
>>>Is there a performance limitation on this configuration (by requiring 
>>>VLAN 1 to do all of the routing between the interfaces and the rest of 
>>>the internet). Peak aggregate traffic is > 1000Mb/s, typical traffic is 
>>>around 300Mb/s right now.
>>>
>>>There is a strange problem that appears occassionally, and is not 
>>>predictable. The problem is the hosts are not able to trace through the 
>>>router. Traces show the router IP at hop 1, and then stars from there 
>>>onwards. Traces in from the internet work fine all the way to the host. 
>>>TCP connections (telnet to the host) do not even connect, but work fine 
>>
>>>from the CPE router. This obviously causes the bulk of the problems.
>>
>>>I am _wondering_ if this is a broadcast problem as broadcasts might not 
>>>be being re-sent down each interface, and since there is the legacy 
>>>problem with the addressing, a simple broadcast helper might not cut it.
>>>
>>>I don't want to configure a bridge group because the total traffic 
>>>exceeds a single link, and Etherchannel doesn't work because each port 
>>>goes to a different aggregation switch.
>>>
>>>My understanding is that this configuration should work, while being 
>>>less than optimal. Further, the configuration did work, but has recently 
>>>begun showing issues for the customer, possibly correlating to an 
>>>increase in traffic flows around the Holiday season.
>>>
>>>Is there a big difference between VLAN 1 and one of the others? The 
>>>example I saw on the Cisco web site showed VLAN 1 being disabled, so I 
>>>don't know if the solution is that simple or its something more 
>>>problematic.
>>>
>>>Thanks in advance,
>>>
>>>DJ
>>>
>>>
>>>
>>>
>>>_______________________________________________
>>>cisco-nsp mailing list  cisco-nsp at puck.nether.net
>>>https://puck.nether.net/mailman/listinfo/cisco-nsp
>>>archive at http://puck.nether.net/pipermail/cisco-nsp/
>>>
>>>
>>
>>
>>_______________________________________________
>>cisco-nsp mailing list  cisco-nsp at puck.nether.net
>>https://puck.nether.net/mailman/listinfo/cisco-nsp
>>archive at http://puck.nether.net/pipermail/cisco-nsp/
> 
>