From: Arne S. <ar...@rf...> - 2025-08-06 05:34:19
|
Am 06.08.25 um 05:46 schrieb Jon Chiappetta via Openvpn-devel: > Hi, > > This is my first time posting to a mailing list. I worked on a possible > performance proof-of-concept modification to the OpenVPN source code. > The code is in rough shape as I just wrote it this long weekend but I > wanted to see if anyone had any interesting feedback for such work. > > Blog Post: https://fossjon.com/2025/08/05/modifying-openvpn-source-code- > to-allow-for-bulk-reads-max-mtu-and-jumbo-tcp-for-highly-improved- > performance/ <https://fossjon.com/2025/08/05/modifying-openvpn-source- > code-to-allow-for-bulk-reads-max-mtu-and-jumbo-tcp-for-highly-improved- > performance/> > > Max MTU which now matches the rest of your network clients (1500 bytes) I have no idea what this means. OpenVPN always supported different MTU sizes and MTU 1500 is standard. > Bulk reads from the tun interface (6 reads) > TCP connection protocol only (single jumbo write transfers) > Performance improvements now allows for (6 reads x 1500 bytes == 9000 bytes) So basically, from your explanation the main thing is to try to write/read multiple packets in one batch on the TCP socket instead of just reading/writing a single packet. I can see that this improves performance. Do you have any idea how much this actually improves performance? Your blog post shows a speedtest with 500 MBit/s but that is not actually saying much. In my own test setup, the current OpenVPN is capable over a Gigabit, a before/after from your test setup would nice to see. > POC Diff: https://github.com/stoops/openvpn/compare/main <https:// > github.com/stoops/openvpn/compare/main>…mods Might be good enough for a POC but from quick glance it looks like there needs to cleanup make this an actual commit. > > If this is the wrong place I apologize in advance, > Thanks for your time, No, this is the right place. Arne |