We have 2 speedtest servers (RHEL 6.7) sitting on 10GB links with a 10GB pipe between the two. The servers are separated geographically by about 256 miles but traceroutes look very good, with 4 hops and and avg RTT of 5ms during an Ookla speedtest and no packet loss on either side. We can get very close to the theoretical max of the link with iperf:
[ ID] Interval Transfer Bandwidth Retr
[ 4] 0.00-7.00 sec 3.68 GBytes 4.51 Gbits/sec 0 sender
[ 4] 0.00-7.00 sec 3.68 GBytes 4.51 Gbits/sec receiver
[ 6] 0.00-7.00 sec 3.67 GBytes 4.51 Gbits/sec 0 sender
[ 6] 0.00-7.00 sec 3.67 GBytes 4.51 Gbits/sec receiver
[SUM] 0.00-7.00 sec 7.35 GBytes 9.02 Gbits/sec 0 sender
[SUM] 0.00-7.00 sec 7.35 GBytes 9.02 Gbits/sec receiver
We see these kinds of speeds regardless of which server is running iperf in client or server mode. However with the Ookla speedtest software the best we can get is about 800Mb down and 4-5Gb up. So far we have been unable to explain this discrepancy and it seems the iperf results suggest an optimally configured network. So we are now looking at the software. I made a change to the Ookla settings file and put it in "gigabit" mode which brought the upload to 7-8GB consistently, but the download remains unchanged and still never gets above 1Gb. Our network team also reports no errors on any or the router interfaces. My Ookla settings file looks like this:
<configuration connection="gigabit" />
<!-- NetGauge settings:
activetests - test list from speedtest,latency,packetloss,firewall
-->
<socket-download testlength="40" initialthreads="500" minithreads="500" maxthreads="500" startsamplesize="500000" maxsamplesize="1000000" />
<socket-upload testlength="40" ratio="1" threads="500" mintestsize="250K" maxchunksize="100M" maxchunkcount="100" disabled="false" />
Bear in mind those settings have not been static as the individual parameters have been changed many times during testing, although I think the one that seems to make the biggest difference is changing the connection type. (lan, broadband, fiber and gigabit). Has anyone ever experienced this kind of asymmetry with Ookla?
