Ethernet Switches & Routers

Reply
Occasional Contributor
Posts: 6
Registered: ‎04-08-2015
Accepted Solution

MLX4 - cam-partition size

[ Edited ]

Hello,

 

Is the logical size dynamic in the sense that lets say we'd have 450k routes injected in IPv4 and 50k routes in IPv4 VPN? If yes, would that mean that, out of the total 768k CAM memory on those units,  I could configure the following:

 

no cam-partition profile ipv4 <- reverting to the default

enable-acl-cam-sharing
cam-mode ip dynamic
cam-mode ipv6 dynamic
cam-mode ipvpn dynamic

 

The above grants of course that I'll discard a fairly large volume of routes received by BGP before they get injected in the RIB + add default routes to be able to send out to those destinations, will I be capable of running near 75% full BGP tables and make use MPLS functionality without killing the cam?

 

I can't upgrade for a while and I'm applying some ugly duct tape. I've got sflow analysis going for the past few months and I know of most of the destinations I rarely/never send traffic to (makes it very easy to drop a lot of routes).

Occasional Contributor
Posts: 6
Registered: ‎04-08-2015

Re: MLX4 - cam-partition size

I think I got my own answer in the documentation... ooops :)

Occasional Contributor
Posts: 6
Registered: ‎04-08-2015

Re: MLX4 - cam-partition size

[ Edited ]

The below applies to those who can't upgrade for a few weeks / months.

 

For those of you who run MLX-4 and receive full bgp tables, the only trick (not ideal) is as follow:

- add default route per core router facing upstream IP transit

- create a prefix-list with all the prefixes you don't really push traffic to (eg: 1.0.0.0/8 le 24)

- add a deny route-map to your bgp peer-group inbound

- watch the routes get dropped, add as much as required to free up the RIB memory

 

Given you're using ipv4, you have to look at your cam-partition usage. This is important as you'll know per slot how much cam entries are used. Take the new profile specs and calculate how much space you'll have left per protocol / per slot when adjusting to the new cam. Never ride beyond 90%, that's how you end up with a full cam and a box of pain. Make sure you at least have 15-20% available in your choice. If some cards ports are too heavy on the cam, split them up on multiple cards.

 

When selecting the cam, if you have dynamic statements on cam-mode, if you pick a non-logical one (static), the statements are pretty much useless from my findings.

 

This of course is in no way, shape or form a long term fix. This is just to get you by until you upgrade. Always be careful of what you merge with network wise, you might get more routes / apply more ACLs at a moment's burst because sales decided to push a new 10G  customer. Beware and keep a very very very close eye on the logs. Make sure you have proper syslog pushed out / proper monitoring of the MIBs before you go ahead with the above cam-partition change.

 

Join the Community

Get quick and easy access to valuable resource designed to help you manage your Brocade Network.

vADC is now Pulse Secure
Download FREE NVMe eBook