[c-nsp] SUP720's memory, looking at options..

Howard Leadmon howard at leadmon.net
Mon Jul 4 01:43:10 EDT 2016

  I knew with the 720-3BXL's I was running, that eventually the TCAM would
become an issue, but it seemed like I still had a little bit of breathing
room left.   Then I saw the chatter here about the RAM on the RP exhausting
before the TCAM, so went peeking at the switch after reading an earlier
thread.     Sure enough, though TCAM was starting to get full, to my
surprise when I looked at memory, it was at 92%, so even closer than the
TCAM by far to exhaustion.  


I know I can't just up the RAM on the board, so that now leads me to wonder
what are reasonable options to resolve this before it becomes a very real
and big problem.   First let me say, compared to many here we are small
guys, we have a limited budget, and our 6509 has served us well for a great
many years, I think it's about to pass the 5yr uptime mark.   We have 2-3
full feeds as uptime is important, and we also peer at the Equinix IX, so
have a bunch of additional peering sessions.


So I started looking around, reading various posts here, and talking with a
few people I know on how to resolve the 720-3BXL running out of gas before
it's a serious problem, but without having to mortgage my home and first
born to do it.  Some of the suggestions I have seen are as follows.


Replace the SUP720-3BXL's with RSP720's, as though the TCAM is still
limited, it will remove the issues of RAM being a problem, as the board can
handle more RAM, plus the added bonus of a much faster CPU that I am sure
would be a plus with BGP.  At the cost of a 7606 chassis, it would be no
biggie to replace it, heck it would save some rack space.   It would also be
a fairly cheap upgrade, with re-usability of my GE and 10GE boards.  Looking
at the DFC's on the interface boards, it looks like there is plenty of room
left in the 1G RAM on the boards.


Another option I was looking at was adding in an ASR 1004 to handle the BGP
connections, and letting the 6500 remain to handle all the local network
ports.  It looks like using something like a 1004 with RP2 and something
like an ESP 20 or 40 would do quite well, and again seems within the range
of affordability on the used market.   Looking at the datasheets it appears
they can easily handle 4 million routes, so that should last a long while.


The last suggestion someone tossed out to me was that I shouldn't discount
the GSR, something like a GSR 12404 or 12406 with the PRP-2 processors, as I
am told they can handle a lot of routes so shouldn't be a problem for the
future.   I spent some time digging around for info on the GSR, but heck if
I could find anything that listed what the BGP route capacity of the suckers
were.   Any hints on this would be most welcome.


In closing, my needs routing BGP wise, would be something with a few GE
ports, and a few 10GE ports.   As I am guessing I will tie back to the local
switch over redundant 10GE links, and we have a few GE transit links, and
some fractional 10GE links as well.   I see no need for anything at all like
40GE or anything huge like that, so our bandwidth isn't massive.  


Any ideas or opinions as to what would be a good option, or bad option would
be most appreciated, and happy to consider any ideas outside of what I have
already looked at.






Howard Leadmon 


More information about the cisco-nsp mailing list