46 minutes ago, Bertel said:

P.S.: The Zonealarm impact is perplexing. The 10GBE NICs and network segments were excluded from Zonealarm. Also note the speed difference between TURNING OFF Zonealarm and REMOVING Zonealarm. 

Many times AV software will insert their sniffing software in the network stack, if it's not well optimized, it can gum up the works for everything. Typically stopping the software does nothing to remove their filter from the network.

 

I always recommend scorched earth removal to test the true effects of a specific AV / security software.

57 minutes ago, jonathanm said:

I always recommend scorched earth removal to test the true effects of a specific AV / security software.

True. However, as much as going from 2 Gbits to 3  may feel like an improvement, let's not forget that Windows should do much better, and as my tests show, a previous version did.  Also, Kaspersky did  not affect the speed.

1 hour ago, Bertel said:

True. However, as much as going from 2 Gbits to 3  may feel like an improvement, let's not forget that Windows should do much better, and as my tests show, a previous version did.  Also, Kaspersky did  not affect the speed.

I think there could be a couple of variables not accounted for in your quest for speed. Having tested on Linux that eliminates the hardware path being an issue. However, have you considered Windows drivers? You might want to experiment with the driver provider and the driver version. I also agree with @jonathanm that any third party software installed on your Windows host is suspect. The only way to test that, is with a clean Windows install without third party applications installed. Windows A/V software and firewall software have the potential to severely impact your performance.

12 hours ago, argonaut said:

I think there could be a couple of variables not accounted for in your quest for speed. Having tested on Linux that eliminates the hardware path being an issue. However, have you considered Windows drivers? You might want to experiment with the driver provider and the driver version. I also agree with @jonathanm that any third party software installed on your Windows host is suspect. The only way to test that, is with a clean Windows install without third party applications installed. Windows A/V software and firewall software have the potential to severely impact your performance.

Thanks for the advice, but as stated in prior posts, several windows drivers, even different NICs were tested without any change. Furthermore, as mentioned several times, the 3 Gbps were reached with a completely new install of current Windows 10 2004, whereas a clean Windows 10 1909 reached around 6 Gbps, and Ubuntu-to-Ubuntu reached ~9.4 Gbps, all on the same hardware, all without any AV software. Zonealarm definitely made bad readings worse, but the main culprit appears to be Windows.

I've also had troubles with tuning 10Gbe (Intel X550-T2) on Windows 10 (2004). And for now i've managed to solve them. Here is a list of a noticable things that helped me to get 9.8 Gbits/sec send and ~9Gbit/sec receive on windows 10:

 

1. Run tests for incoming and outgoing traffic. Use approved testing tools. I had a wrong test results with wrong iperf versions for windows. The one that worked for me was 3.10.1  the "official" 3.1.3 from iperf.fr didnt saturate incoming bandwidth and reported only 4.2Gbit/sec. 

2. Install latest driver from vendor's site. The built-in windows driver didnt perform well (about 6Gbits/sec)

3. Enable RSS queues and set it equal to the number of physical cores or less.

4. disable all power save features. Set power plan to  performance.

5. disable interrupt moderation

6. enable jumbo frames (9k) on every network device on path

7. Ensure TCP Window auto-tuning is set to normal 

8. The NIC should be plugged in in at least PCIe 8X slot

9. Enable offload features

10. Disable AV/Firewall and any other service that filters network traffic.

 

Edited July 23, 20214 yr by yoshi kakbudto