Just so someone else doesn’t go through what I’ve gone through the past couple days…
CCR1036 with 2 full table BGP sessions and a bunch of static routes and OSPF has crashed multiple times.
First CCR was looked at by Mikrotik and they said it had bad RAM. Replaced it with a second CCR.
Second CCR running 6.18 crashed within 12 hours of being deployed with the same configuration.
I’m ready to throw all of this CCR garbage in the trash.
running bgp with 3 peers, 1gbps bandwidth running, version 6.19. Running from 17days uptime (rebooted because of upgrade). The only problem i faced was bgp filters not behaving the way it should. Memory consumption increasing. It started up with 34xxMb free and after 17 days its 3260Mb free. I’m checking memory consumption regularly it is increasing at slow pace. On my earlier x86, ver 5.2 the memory was rock solid of months of uptime. Still early days to deduce anything.
One thing is for sure it doesnt reach anyway near their advertised benchmarks (using normal operations. cannot go beyond 1Mpps approx)
3 full routing (upstream) peers + 2 route reflectors + clients peers = 49 day uptime (RouterOS update).
I’m not saying it’s “good”, but you can live with it.
My issues are mainly on performance. It takes about 5 minutes to load a full table and can take more than an hour if two peers are loading simultaneously.
It would be nice if the BGP processes could be shared among the CPUs. Instead I have a 100% CPU and 35 5% CPUs
In another thread, one of the staff (sergejs i think) said that current ROSv7 tests have 15 million routes installing in 15 minutes (so 1mill routes / min) which seems a big improvement.
Ram for the CCR series is pretty inexpensive. I had an 1036-8g-2s±EM model show similar symptoms. Replaced the 16G with one 8G stick (http://www.neweggbusiness.com/product/product.aspx?item=9b-20-139-976) and its been up for 34 days since pushing a few gb/s average with 3 full tables.
My recommendation, try replacing the ram and upgrade to 6.19.