[netperf-talk] HELP:Regarding iANS test using netperf
Rick Jones
rick.jones2 at hp.com
Wed Aug 23 11:10:22 PDT 2006
> As part of our project we needto use netperf tool to check the iANS's
> performance.Here comes the topology.
>
> IO SERVER-----------(two kilomega twisted pair --------CISCO4506
> (with large storage) wire,bonding by iANS) | | |
> | | |
> | | |
> | | |
> One of client node------(single kilomega twisted pair)--------- | |
> One of client node------(single kilomega twisted pair)----------- |
> . . . |
> . . . |
> . . . |
> One of client node-----(single kilomega twisted pair)--------------
>
>
> *******************************************************************
>
> First I test one IO server to one client node,the result as below
> (IO server set to be test-server)
> [root at rack64 src]# netperf -H 192.168.77.8 -l 60
> TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to xinnode3
> (192.168.77.8) port 0 AF_INET
> Recv Send Send
> Socket Socket Message Elapsed
> Size Size Size Time Throughput
> bytes bytes bytes secs. 10^6bits/sec
> 87380 16384 16384 59.99 904.44
>
> That looks like OK.
>
> But when I try to use two client nodes 'netperf' the test-server in the
> same time ,the results comes like this :
>
> [root at rack61 netperf-2.4.1]# netperf -H 192.168.77.8 -l 60
> TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to xinnode3
> (192.168.77.8) port 0 AF_INET
> Recv Send Send
> Socket Socket Message Elapsed
> Size Size Size Time Throughput
> bytes bytes bytes secs. 10^6bits/sec
> 87380 16384 16384 59.99 470.13
>
> [root at rack60 netperf-2.4.1]# netperf -H 192.168.77.8 -l 60
> TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to xinnode3
> (192.168.77.8) port 0 AF_INET
> Recv Send Send
> Socket Socket Message Elapsed
> Size Size Size Time Throughput
> bytes bytes bytes secs. 10^6bits/sec
> 87380 16384 16384 60.00 473.04
>
>
> ************************************************************
>
> The IO server suppose to be a powerfull one which can provide 2000
> bits/sec, but according to the test result,it just about 940 bits/sec.
>
> what can I do ?
You need to look into the specifics of the algorithms used by iANS on
either end to do the bonding. What fields of packet headers does it
examing when determining on which of the links it should send a packet -
in each direction.
For example, if the Cisco side is looking only at destination IP
address, then it doesn't matter how many connections are coming to that
one IP address, they will all be put over the one link.
Most bonding/aggregation/trunking solutions, at least those in their
right minds :) will not spread the packets of a given "flow" across
multiple links - otherwise the packets in that flow could be reordered
and that would be bad. So, if the device(s) in question define a "flow"
as the destination IP...
rick jones
More information about the netperf-talk
mailing list