Closed DMarby closed 5 years ago
Thanks for the PR! I had suspected that I implemented this code in a very naive way. I'll take a closer look on Monday, but could you please share the performance numbers you are seeing before and after this patch? Thanks!
When testing with 20.000 peers, running on an Intel Xeon E3-1230 v6, I'm seeing the Device(name)
call take 5-10 seconds to finish previous to this patch, with this patch it's near instant (< 300ms)
Hi,
Thanks for your awesome work on this library!
While using the library against wireguard devices with a large amount of peers, I noticed a big slow-down and a significant increase in CPU usage. I narrowed it down to the code for merging the peer information sent across multiple netlink messages, so this PR contains an optimized version of that.