[f-nsp] Packet loss on MLXe-4

[Frank Bulk (iname.com)]

Thanks for the feedback received.

We just completed a scheduled maintenance window with Brocade tech support
where both issues were resolved.  

When we inserted the MLX into the ring we saw IP video traffic at some CES'
on the metro ring drop from ~700 Mbps to ~170 Mbps.  The Brocade tech
noticed that we didn't have the "router pim" command configured at the
global level on the MLX.  We applied that command, but it didn't resolve the
IP video issue. But while the Brocade tech was poking around, my colleague
looked at the only other MLX on the ring and noticed that we didn't have
"multicast flooding" configured on the two VLANs carrying the (multicast) IP
video traffic.  He applied that to the two IP video VLANs on the problem MLX
and the IP video traffic levels restored back to normal.  We learned from
Brocade TAC that without this command the LP's CPU is impacted, and so to
protect it some traffic is thrown away.  If we had run "show cpu lp" we
might have seen the LP's spiked traffic levels, but alas, we only ran "show
cpu".  On the CES' this "multicast flooding" feature is on by default, but
with the MLX you have to explicitly configure that.  

We also moved the inter-border router link from eth 1/2 to eth 2/2 and that
was just fine now, too  We theorize that the LP's CPU was hit so hard with
the multicast traffic that while the BGP and OSPF packets were properly
processed (probably because they have higher priority), unicast and other
traffic was affected.  

Frank

-----Original Message-----
From: foundry-nsp [mailto:foundry-nsp-bounces at puck.nether.net] On Behalf Of
Frank Bulk
Sent: Tuesday, September 17, 2013 3:17 PM
To: foundry-nsp at puck.nether.net
Subject: [f-nsp] Packet loss on MLXe-4

We inserted an MLXe-4 into a Brocade ring this morning and it didn't go so
well.  We saw most IP video traffic (1.3 Gbps) not flow through, and when we
moved an inter-border router link from eth 1/2 to eth 2/2 we saw significant
packet loss on other Internet VLANs (all on eth 1/x).  Moving it back
resolved the issue.

We have two SFMs and have two four-port 10G cards, one is a NI-XMR-10Gx4 and
the other is a BR-MLX-10Gx4-X.  We also have a BR-MLX-1GCx24-X in slot 3
that we're not using, yet.  We're running 5.4.0d and the FPGAs are at the
correct releases.

The ring goes in one 10G port on the one 10G card and through another 10G
port on the other 10G card.

Are there any compatibility issues between the two cards?  Placement
requirements?  Any obvious reasons why we experienced all that traffic loss?

Regards,

Frank



_______________________________________________
foundry-nsp mailing list
foundry-nsp at puck.nether.net
http://puck.nether.net/mailman/listinfo/foundry-nsp