We have installed RB1100AHx2 on BGP peer
But the board does not provide the stated capacity.
Even at speeds of 400-500mbit and 40-50K pps 90-98 percent CPU utilization.
Firewall is on
Conntrack is off
Raised 3 vlans and BGP to one peer
The documentation to this board claimed such data
Routing 2334.72mbit 570000pps 3643.2mbit 300000pps
How to achieve this capacity?
It’s UDP traffic I think…and large frame sizes.
I’ve managed to get about 1.2-1.4 Gbps thoughput of TCP traffic.
Bonded 2x 1Gbps ports.
No, the size of packets is standard.
UDP or TCP, what’s the difference for the bandwidth? )))
What would the experts?
You quoted tests that ran with 512 and 1518 byte frame size, and UDP traffic. Is your traffic also UDP and at those frame sizes?
The main volume of traffic on the packet size between 1024-1518
The main traffic is UDP (torrent clients), then TCP
Will it help to us change the CPU frequency to 1333MHz?
Can to us help option “switch”?
and one more thing
If I turn on the conntrack - then the rate drops to 70-100 Mbps
cpu load to 100%
and begin great losses.
On this board there are two switch.
Switch 1 - 6-10 Port
Switch 2 - 1-5 Port
If I make a connection uplink and downlink for example on switch1, then it will improve the performance of board?
best performance will be if you will use one port on one switch, other port on other switch.
I have done as well, uplink on port 10 and downlink on port 1.
So the options of “switch” will not help here?
Second place on volume of traffic occupied by packets ranging in size from 64 to 127 bytes.
Whether it can affect the decreases bandwidth the device?
Did the bonded device will increase the capacity, if the overload is on the CPU?
Have you checked?
/tool profile
Yes The documentation clam 2334.72mbit 570000pps 3643.2mbit 300000pps with firewall on
but if Filter, Mangle etc .. are empty no Queue no BGP
but not sure??? there is so little info about that
I think test are made on most simple routed traffic
I have same experience with RB1200
All those tests are done with minimal configuration (for example routing = 2 IP addresses with connected routes), and should be used only for comparison of RBs between themselves.
So that in case you determine that you need more powerful router you can choose right one.
Yes checked, the main load is on the queuing, then the firewall, ethernet and routing.
Queuing are in default (only-hardware-queue, kind - none)
On the firewall only one rule.
Same here, queing eating 10-15% cpu, but I don’t have queues enabled.