[c-nsp] carving TCAM on Engine 3 linecards due to exhausted IPv6 TCAM regions

[Per Carlson]

Hi Jan.

> after some flux of IPv6 routes I got on my last old GSR the error
> message for each line card:

We experienced that as well today.

> Issueing a "sh controllers frfab alpha tcam carve" shows me an
> utilization of 50-100% of any RX_IPv6_* region while all other regions
> have 0 to 5% utilization. I've no idea what the other regions exactly
> are used for except Netflow.

The others are for ACLs, MQC, policy routing etc. (from the ouput
above "WARNING: Recarve may affect other input
features(ACL,CAR,MQC,Netflow)").

> Does anybody have experience in re-carving this TCAM?

We have had an discussion with Cisco about the topic today. If you
lookup CSCsx52504 you will find an explanation of the default TCAM
carvings and how to change those.

> Do I need to issue this command for each prefix lengh where a region
> exists for or is there some magic command?
>
> #hw-module slot 1 tcam carve RX_IPv6_144b_REGION ?
>  <0-128>  Prefix length
>
> #hw-module slot 1 tcam carve RX_IPv6_144b_REGION 128 ?
>  <0-100>  percentage of TCAM

There is a magic command. You only allocate a percentage for IPv6 and
it will be divided per prefix length automatically.

>From the bug description: "For the purposes of IPv6 scalability, the
RX_IPv6 regions are the ones we're interested in. Initially, all IPV6
cells are placed into the /128 region. As we add other mask lengths
into the ipv6 routing table, we will "move" blocks of the TCAM into a
new region that corresponds to the mask length for that prefix.".

Further on:

To support more IPv6 routes in hardware:
Get current TCAM usage with: show controllers ISE tcam
In config mode, reallocate TCAM regions e.g. reallocate Netflow TCAM to IPv6

hw-module slot < num> tcam carve rx_ipv6 < v6-percent>
hw-module slot < num> tcam carve rx_top_nf < nf-percent>

Verify with show command that sum of all TCAM regions = 100%

Reload the linecard for the new TCAM carve config to take effect

-- 
Pelle

RFC1925, truth 11:
 Every old idea will be proposed again with a different name and
 a different presentation, regardless of whether it works.