RB1100AHx2 Performance

I have a RB1100AHx2 (v5.12) with below configuration

  • Running BGP, 6 x iBGP peers, 7000 routes in routing table
  • Connection tracking disabled, no firewall, no bridges, no queues, no NAT, no vlan, no bonding
  • Two lines of mangle rules for routing mark (matching from 20 lines of address-list)
  • Traffic level about 400-450Mbps
  • PPS about 70-100k (tx/rx combined), avg packet size (tx rx combined) is about 600-700 bytes
  • CPU load 5-min avg 60%, Winbox shows 56-70%
  • CPU profiling shows ethernet 12%, firewall 20.5%, idle 49%, queuing 17%, others 1.4%

I’m seeing 1-2 icmp packet losses every 1000 packets. I have other x86 based routers using ROS 5.x with more complex configurations such as queues, firewalls etc. but none of them are dropping any packet.

Anyone else having this issue? Any idea why it drops packets?

if you have no firewall and no queuing, why the profiler report your cpu usage ?
try to update ros.

Well my guess was that the mangle rules counted as firewall. However I have no idea why I see cpu usage for both queuing and ethernet. There is only default only-hardware-queue for each gige interface and that is it, no simple/htb queues configured at all. Tx/Rx drop counters are all zero. IMHO the router has plenty of idle cpu and shouldnt discard any packet.

I’ve read ROS changelog from 5.12 to 5.20 latest stable but none of them had changes to firewall, queues etc but I’ll upgrade to the latest version tonight.

I have tried RB1100AH but not x2
but In reality rb1100AHx2 or any rb1000-1200 series CPU is not like x86 CPUs (xeon, i5, i7)
so dont expect miracle

400-450Mbps for rb1100AHx2 +BGP CPU 50-70% is good performance

about ur post I rely dont understand but if u using routing or NAT connection tracking is essential
enable connection tracking it shouldn’t be so much problem for CPU
correct me if I m wrong.