[c-nsp] Cisco 4000 series (4461) as a BGP router?

[Saku Ytti]

On Mon, 28 Oct 2019 at 16:55, Gert Doering <gert at greenie.muc.de> wrote:

Hey,

> But this should not be an excuse for a vendor not publishing basic
> things like "the forwarding hardware on <this generation foo> can
> handle 4.13 trillion IPv4 entries, unless you use IPv6, in which case
> it goes down to only 22.000".
>
> FIB is fairly easy here.

In MX for example this is multidimensional answer. Same memory is used
for many things and having less A means you can have more B.

Here is one MX:

IMPC0(r33.labxtx01.us.bb-re1 vty)# show jnh 0 pool
                Name     MemType            Total         Used
(%)    Free       (%)

            Next Hop     EDMEM               2621440       2001912
76%       619528  24%
                         Bulk_DMEM           1048576        458976
43%       589600  57%
                         IDMEM                507904        391633
77%       116271  23%

            Firewall     EDMEM               2621440         35977
1%      2585463  99%

            Counters     EDMEM               6815744       3195230
46%      3620514  54%
                         IDMEM                 12288          3798
30%         8490  70%

         Services NH     Bulk_DMEM           2097152         16397 <
1%      2080755 >99%

                LMEM     LMEM                    128           128
100%            0   0%

                HASH     EDMEM               3937792       3937792
100%            0   0%
                         Bulk_DMEM           1675008       1675008
100%            0   0%

              ENCAPS     Bulk_DMEM           4259840       4259840
100%            0   0%

          UEID_SPACE     Bulk DMEM           1048576           129 <
1%      1048447 >99%

   UEID_SHARED_SPACE     Bulk DMEM             65536             2 <
1%        65534 >99%


Now either you test, or you acquire understanding what does EDMEM,
DMEM, IDMEM, LMEM mean and how does it apply to your scenario. And
once you upgrade platform you'll do all that again.

> "How many full table feeds will fit into the RIB" is more complicated
> (as it depends on more factors than sheer "number of prefixes" but
> also paths, attributes, churn = CPU usage, etc.) - but even there a
> rough number is easy to publish.

Perhaps it seems more complicated, because you understand it better.

-- 
  ++ytti