Anyone got some real world experience with the CCR1036 and large routing tables? I’m expecting to take on 2x global tables + probably another 50k-100k routes via various local peerings, so could be approaching 900k-1M routes, anyone pushed a 1036 that far?
I know BGP processing is limited to a single core so when a global table peer drops its gonna take a bit to recover, but just pure performance wise holding that many routes will be interesting to hear if anyone installed it in this situation?
we run a similar config with two full feeds plus around 100k routes learned via several different peers (15 total) on a CCR1016. The routes take around 650MB of memory.
We’ve been running two CCR1036-4S+'s since the beginning of this year. One of them has 4 transits which all send out full tables, the other has two transits and does public peering (~ 100 peers). All BGP and no problem at all. Memory at 30%. It takes about 3-4 minutes to load up all the tables after a reboot.
I do however recommend running at least two of them and setting up some kind of redundancy (e.g. we are doing VRRP), so you don’t suffer outages whenever a router needs to reboot and reloads all of the BGP tables.
We use 2 CCR’s in a production environment replacing 2x vyatta systems. We receive full detailed tables from multiple carriers on each router, in v4 and v6, and take peering carrier routes from about 5 peers on a local IX. So far, no complaints. I can confirm that the table reload time can be anywhere between 1-3 minutes though.
6.12 so far - probably 6.13 by the time i implement them. Running them in dual failover so will be staging upgrades at a pretty decent rate.
Have played around with a quick hack generated ~100k routes and suddenly dropped/added peers etc. Winbox becomes unresponsive for some time but ssh still runs pretty smooth. I tend to prefer doing major changes on CLI anyways (tend to write an .rsc and run as an import in safe mode)
They are only in testing at the moment but I have 2x 1036-8G-2S+'s connecting to 2 switches each using 2 port 802.3ad, bridged with RSTP for redundancy in switching, iBGP, some test eBGP, OSPF for IGP / nexthop lookup, chucking around ~100k routes. Haven’t had them reboot, but also not done a heap of data.
Because the whole core is being replaced with new equipment we will have easy rollback by simply plugging the old stuff back in, so we will be implementing and leaving the old gear racked for a little bit before we consider it successful.
edit: SFP+ ports will be used for Megaport 10G connectivity also, but so far just have them connected to each other to test.