Network throughput question..

levensailor

I have the following:
client<->3560 SW<> Cisco Small business SW<>server
everything is connected via 1GB links and in the same vlan

Running an iperf test from client to server i'm getting around 500kbps
Is this reasonable?

Would a client on a different vlan run considerably slower, judging the 3560 had to route?
Would connecting the server directly into the 3560 make it considerably faster?
Would running an LACP link between the servers NICs and the small business switch make any difference?

Would the fact i'm using a small business switch hinder me in any way, considering it is a gigabit switch?

mohamedshajid

Why don't you configure the bandwidth? Sorry i'm a newbie but i think it should be case of bandwidth that you'd provided

networker050184

You should be able to get more than 500kbs through routing or not and no, you don't have to configure the bandwidth.

Zartanasaurus

500kbps is very unreasonable for an iperf test. You should see the entire 1Gb. The answer to your questions are unknown. You're going to have to run several iperf tests to find the bottleneck. Connect two clients to the 3560 and run iperf. Do the same thing on the small business switch. If that comes up okay, run an end to end test using different clients and servers than you did in the first test. Maybe the server has a bad NIC.

Check the show interfaces on the 3560 and SB switch and see if you're getting a lot of errors on one of the ports.

f0rgiv3n

With iPerf you should attempt to increase the TCP window size that will give you a higher throughput report. Give it a shot . I believe the switch is -w #

I'd also like to add this link:
https://kb.doit.wisc.edu/wiscnet/page.php?id=11779
It shows exactly what you are seeing and what I have seen in the past. iPerf doesn't necessarily use the Window size that the OS is using.

Another option would be to use UDP instead of TCP.
http://opentodo.net/2012/08/29/test-network-performance-with-iperf/
Scroll to the bottom of that one, it shows that in order to use udp you use -u and -b # to specify the amount sent

deth1k

Window size is only relevant for WAN circuits where round trip is high, so increasing WS will increase the throughput. In this case its a flap LAN. Please advise what iperf commands did you use? If it was UDP based test i suggest specifying bandwidth for the test:

iperf -c x.x.x.x -u -b 1000M (client)
iperf -s -u (server)

f0rgiv3n

deth1k wrote: »

Window size is only relevant for WAN circuits where round trip is high, so increasing WS will increase the throughput. In this case its a flap LAN. Please advise what iperf commands did you use? If it was UDP based test i suggest specifying bandwidth for the test:

iperf -c x.x.x.x -u -b 1000M (client)
iperf -s -u (server)

Interesting info, thanks. What's weird is I have seen iperf give me differing throughput even on a switched(L2) network. I think it selects a small window size by default. Maybe I don't understand why it would only apply to WAN circuits.