I have an on-premise site with mainly Windows resources and an up-to-date Forti firewall as the central communication device. This site is connected through a partner interconnect created according to Google's design, using 2x10gb lines with around 32ms delay. From Google's side, I use cloud routers, shared VPC, and resources in projects. All the Windows OS versions I've tested are Server 2019, 2022, 2016, and Windows 11.
In Linux to Linux communication, there are no problems - good bandwidth and no packet drops. However, in Windows communication, I only get up to 50 Mbps when measured with iperf, and the overall communication is bad. I believe this is connected to the TCP window size. When I try to increase the window size in iperf, the first line shows good bandwidth, but it immediately drops again to around 47-50 Mbps.
I've tried different combinations of ECN, scaling, and auto-tuning with no success.
I've reviewed network design.
Investigated the on prem network environment.
Investigated that all services in Google incl. vlan attachments are not limited. In fact, I only made things worse with the Windows TCP setting. Any ideas would be appreciated.
In the screens are test between linux,linux where i get almost the max for the network card.In the other is between windows-windows , windows linux where i get at most 50.linux-linux
iperftest output to the question? Try both ways, Win client / Linux server and Win server / Linux client. Just edit it, and paste the output enclosed in code brackets{ }to preserve the formatting. In general it's always preferable to add info or give clarifications and answers to questions by editing your initial question rather than answering in comments. That way the question gets more focused, is always up-to-date with your current situation and al the relevant info is readily available.