Network Performance Questions

[idiot_wind]

Hello All,

I am trying to understand the difference in network performance during testing. In addition, I was looking for suggestion on built in Linux tools that would provide more accurate information then the procedure that I am using to test.

 My configuration is as follow, I have 3 servers, all servers are ProLiant DL380 Gen10, 2 servers have NIC cards HPE Ethernet 10Gb 2-port 562FLR-SFP+ Adapters, and 1 server has an HPE Ethernet 10/25Gb 2-Port 631SFP28 Adapter.  These servers are not in production, so they are pretty much idle.  The servers are connected via fibre to an HP 5940 switch that appears to be configured correctly.  All servers are in the same VLAN and on the same switch.

I am using the following to test; I realize that it is not the best way to do so, so suggestions are appreciated.

dd if=/dev/zero bs=10G count=50 | nc hostname 4000

When I run this test between the 2 servers that have the HPE Ethernet 10Gb 2-port 562FLR-SFP+ Adapters installed I see a speed of 230 MB/s.

When running this test between 1 server that has an HPE Ethernet 10Gb 2-port 562FLR-SFP+ Adapter and the 1 server having an HPE Ethernet 10/25Gb 2-Port 631SFP28 Adapter I see a speed of 203 MB/s.

I expected to see higher throughput on the second test as one adapter is a 10/25Gb adapter.

When looking at ethtool on the different adapters nothing jumps out as being misconfigured.  I used iptraf to ensure that I am using the expected adapter to run my tests.

My questions are 1- Should the performance be similar or higher on the second test?  2 – Should I see something closer to 10Gb instead of the high of 230 MB/sec?  What would be a suggested tool that would be more accurate instead of using dd if=/dev/zero bs=10G count=50 | nc hostname 4000?

Many thanks in advance!

frank

[parnassus]

Your method deserves an essential question to be answered first: what are you measuring? your servers' I/O local storage subsystems performance or your network links current maximum throughput?

[idiot_wind]

I am trying to measure network links current maximum throughput.
In addition, I am trying to understand why I would see different results when running my flawed test on different adapter types. When I run the identical test between 2 servers that have the HPE Ethernet 10/25Gb 2-Port 631SFP28 Adapters installed I see a lower throughput rate then when I run the test between 2 servers that have the HPE Ethernet 10Gb 2-port 562FLR-SFP+ Adapters installed. I forgot to mention that these servers are running RHEL 7.6

[parnassus · Accepted solution]

"I am trying to measure network links current maximum throughput."

Really? so why aren't you "hammering" Servers' NIC ports by using a network performance measurement tool such as iperf?

You're currently bringing also servers' local storage performances (bottlenecks) into the "network" equation...thus you're not measuring only the network.

[idiot_wind]

Thank you for your response,

I was able to get permission to download iperf for netwrok testing then once completed with the network testing I needed to do a clean reinstallation of the OS to ensure that there were no artifacts hanging out after the removal of the tool..  It worked well and provided the information that I needed.  Thank you again.