10-28-2011 11:53 AM
attached is a highlevel diagram.
there are multiple fesx switches attached to the vdx cluster.
the MLX MCT is doing all the inter-vlan routing
when servers attached to the FESX ping between themselves or ping their default gateway I see every now and then an increase in latency from submillisecond reply to as high as 20ms.
any reason for this?
I also see this behaviour when pinging between the two mlx's.
10-28-2011 12:38 PM
what about the CPU on the switches?
especially on MLX:
show cpu lp?
or on the FESX:
Is the traffic sometimes flooded as unknown unicast?
10-28-2011 12:50 PM
CPU on MLX is low and does not waiver much. I haven't seen it over 3percent yet.
on the FESX, it is a constant 1percent.
as for unicast flooding. I see the mac in the cam table (for the default gateway) since that is where the first hop is. i don't see this behaviour on servers that do not flow via the MLX cluster, servers on the same vlan even between different fesx's.
Also, in the diagram there are two upstream devices. a load balancer and a mlx. these are connected via a point to point ve and I see the same issue on those links.
interestingly enough from the upstream mlx to an adjacent device, I do not see this behaviour.