x670G2 (trident2) - L2 capacity

[Marcin_Kuczera]

I remember X650 switches with 32k L2 MAC table. However, that table was using hashing algoryth with 8 MAC addresses per bucket so in our environment at over 10k MAC addresses we had a lot of unknow-unicast traffic.

As I know, X670 (128k table) also used some hashing algorythm, as I found some information about enchancements in it to cover more MACs without overrun of particular buckets.
But, I didn't have them.

Now I have X670G2 (288k table, trident2 based) - if I do:
debug hal show fdb
It looks like this switch has a real 288k L2 capacity (normal TCAM).
Hardware learned entries looks like that:
54:e6:fc:b1:cb:b7 3206 00001021 1:36 FALSE FALSE
10:7b:ef??a2:93 814 00001021 1:30 TRUE FALSE FALSE
00:06:31?e7:2f 3851 00001021 1:29 FALSE FALSE FALSE
00:25:90:e5:b7:aa 112 00001021 1:21 TRUE FALSE FALSE
00:25:90:e5:b7:aa 126 00001021 1:21 TRUE FALSE FALSE
c0:4a:00:b7:a0:2f 3067 00001021 1:36 TRUE FALSE FALSE
00:25:90:e5:b7:aa 122 00001021 1:21 FALSE FALSE FALSE
14??20:5b:8e:f5 3888 00001021 1:40 TRUE FALSE FALSE
00:0c:42:7b:ea:41 3846 00001021 1:29 FALSE FALSE FALSE
00:25:22:66:90:31 3063 00001021 1:36 TRUE FALSE FALSE
00:1e:2a:63:5d:7f 3028 00001021 1:36 TRUE FALSE FALSE
00:25:90:e5:b7:aa 121 00001021 1:21 TRUE FALSE FALSE
10:7b:ef??a2:93 814 00001021 1:30 TRUE FALSE
00:06:31?e7:2f 3851 00001021 1:29 FALSE FALSE
00:25:90:e5:b7:aa 112 00001021 1:21 TRUE FALSE
00:25:90:e5:b7:aa 126 00001021 1:21 TRUE FALSE
c0:4a:00:b7:a0:2f 3067 00001021 1:36 TRUE FALSE
00:25:90:e5:b7:aa 122 00001021 1:21 FALSE FALSE
14??20:5b:8e:f5 3888 00001021 1:40 TRUE FALSE
00:0c:42:7b:ea:41 3846 00001021 1:29 FALSE FALSE
00:25:22:66:90:31 3063 00001021 1:36 TRUE FALSE
00:1e:2a:63:5d:7f 3028 00001021 1:36 TRUE FALSE
00:25:90:e5:b7:aa 121 00001021 1:21 TRUE FALSE
Hardware-learned In-use count: 22111
Num of msgs from FDB : -1975822112

So, I'a guessing it is TCAM.

Could someone confirm that ? Or correct me ?
How about X460G2 with 96k L2 table ?

[Stephane_Grosj1]

Newer generation of products have a major enhancement in terms of table lookup size. The user is now allowed to allocate a certain amount of resource to a given lookup table. You have the opportunity to select the configuration that suits you the best (more L2, more L3, etc.), with the following CLI command:

config forwarding internal-tables [l2-and-l3 | more [l2 | l3-and-ipmc]]

You can check your actual configuration (default is l2-and-l3) with the CLI command:

show forwarding configuration

As for the table itself, this is still hash table. There was an enhancement for L2 to offer better hash in some previous EXOS release, I think it was introduced in 15.3.2 on specific platforms, and generalized in 15.6. In particular with G2 platforms, the architecture is different and various algorithms are at play, resulting in a good table utilization.

[Stephane_Grosj1]

The values you gave are maximum. They are dependent of the settings. As for L2, I don't know them off the top of my head, but that's something like 288k, 160k and 96k, depending on the settings. It's all hash table.

[Stephane_Grosj1]

That must be documented in the User Guide.

p.1313 of 16.1 User Guide, you'll find the L3 values.

It doesn't seem L2 is present in the doc after a quick look.

[Marcin_Kuczera]

ok, but let's say that speciification sais:
Layer 2 / MAC Addresses: 288K
IPv4 Host Addresses: 136K
IPv4 LPM Entries: 16K
IPv6 Host Addresses: 48K
IPv6 LPM Entries: 8K

Is it all at once, in the same moment or these are maximum values ?
For X670G2 - how is config forwarding internal-tables supposed to work ? How does it share entire memory ?
I would like to see i'e that in default config there is i.e. 128k entries for L2 MAC, 8k for IPv4 routes, 1k for multicasts etc..
It is still very unclear what are the real max limits in particular config.