Buffer packets in Ios XR AS9k (tcp performance)

[Zemilion18]

Hi! I have 2 Asr9k routers, and the TCP performance in tests such as iperf between servers with 10G interfaces on both routers is somewhat low (500-900Mbps), even with RFC1323 tags. There is a latency of 10ms between the routers due to the distance, so perhaps that is why the TCP performance is not exceptionally high but only regular. However, I don't think that's the cause but an additional consequence. I believe that this is probably a packet buffering problem, the "sh interface X" command does not show me the packets in queue (IOS XR). In the same transport network, I have routers from other brands that I test in exactly the same way with the same servers, and they give me 4Gbps TCP, so something is happening with the Asr9k routers. The dataplane has no problems, only default and OSPF routes, the trunk interfaces are at 400G, and there is 200Gbps of traffic, so the packet count is very high. None of the interfaces have errors, and the modulation is fixed at all points. Does anyone have any ideas? Tests in UDP give the full capacity.

[Julio E. Moisa]

Hi

Have you tried to modify the MTU values or if is the same value on both sides? Do that during a maintenance windows. 

Also check the interface speed and if they are working in full duplex. 

Regards, 

[Zemilion18]

Hi, The 2 servers operate with an mtu at 1500 bytes, and the trunks are enabled for 9 thousand bytes so there are no mtu bottlenecks, a ping with the no fragmentation flag completes without problems between the servers.

[Julio E. Moisa]

Thank you for the update, have you certified the link between the peers? or anything related to the physical layer?

Do you see any error on the interfaces?

Any process with high CPU/memory utilization?

Regards,

[Zemilion18]

There are no errors in the interfaces. I have tested the same physical layer with the same test servers but with routers from other brands, there is no problem there. However, I have carried out these tests without traffic, with the Asr9k there is 200gbps of traffic, so the packet count is high, the npus have a utilization of less than 5%, the cpu in general does not rise above 3%.