Slow throughput Nexus 5k 10GB to 1GB

[s.leyer]

Hello,

We have a Nexus 5020 with a FEX 2248. Server A is directly connected with 10G at the Nexus 5k and Server B with 1G at the FEX. Both servers are in the same Vlan and running with Linux (Ubuntu 11.10 and Scientific 6.1).

By running a bidirectional iperf-test, the throughput from Server B to A (1G to 10G) is 940 Mbits/sec and from Server A to B (10G to 1G) is only 140 Mbits/sec.

The performance from Server A to B is not favorable for e.g. a file server.

Can anybody help to optimize ?

Unidirectional tests give in both directions 940 Mbits/sec.

Thanks

[andre even]

Hi,

I have discovered the same issue in an environmment with 2 x Catalyst 4500-X in VSS.

Regards.

Andre

[Leo Laohoo]

Andre,

Your issue is not "the same".  You've got a 4500X in VSS but the OP is a Nexus.  Two different beast. 

Can I reqest if you create a new thread so it won't confuse other people?

When you do, please also post the output to the following commands: 

1.  sh interface ;

2.  sh run interface ; and

3.  sh counter interface

[3ssheedy]

Hi All

I believe I have the same (or very similar) issue.

Server A connected with 10GB NICs in LACP to Nexus9000 C9372PX

Server B connected with 1GB NICs in LACP to N2K-C2232TM-E-10GE

Both servers running Red Hat Linux

File transfer from 1Gb to 10GB delivers at line rate. All good!

File transfer from 10GB to 1GB starts but soon grinds to a halt. Falls over.

Looks like "bufferbloat" problem. No errors seen on any network or server interfaces.

Wireshark captures on both sides shows high volume of DUP ACK's followed by retransmissions.

Anybody got any simple solutions to this issue?

Does tweaking the "receiver"window size help with this problem?

Thanks

Senan 

[Joseph W. Doherty]

It might. Try sizing the receiver's RWIN for your BDP.