Repository navigation
Replies: 1 comment 2 replies
|
Thanks, this looks like solid work, and the interop suite against kernel MPTCP is a good sign. OpenMPTCProuter tries to offer every aggregation technique available: kernel MPTCP through the proxies, Glorytun, MLVPN, mqvpn's multipath QUIC, and so on. Users pick what suits their links. A userspace MPTCP stack would fit in as one more option, and I'd be glad to add it once it can carry real traffic. Some context on where OMR is now. The VPS limitation from #4229 is mostly gone: the VPS runs our own 6.18 kernel with the BPF struct_ops schedulers, and the router syncs the scheduler choice to it. So the remaining question is whether a userspace stack does better than the kernel there, not whether we can get schedulers at all. What would make it hard to use in OMR today:
The place where it could really help is the VPS: a server that accepts the routers' kernel MPTCP connections and lets us pick the scheduler per connection. Today net.mptcp.scheduler is global, so a redundant or DSCP-pinning scheduler affects every proxied flow. A shadowsocks (or SOCKS) server on mptcp-rs, compared with kernel MPTCP on a 1–2 vCPU VPS with a few routers each driving many concurrent flows, would show quickly whether it's viable. Once the server role handles that load, I'm happy to run it on our test bench. |
Uh oh!
There was an error while loading. Please reload this page.
Hi,
I have recently been working on developing a user-space MPTCP stack out of personal interest,
https://github.com/mp0rta/mptcp-rs
and I thought it might be useful for OMR.
I saw the VPS-side limitations with BPF schedulers discussed in #4229, and I think a userspace implementation could be an interesting alternative.
This approach would allow for more flexible scheduler customization without requiring a custom kernel on the VPS. I am also considering integration with DPDK/AF_XDP, which could potentially deliver performance equal to or better than the kernel-based MPTCP implementation.
Although the project is still in an experimental and I have only conducted interop tests using TUN interfaces, but I was wondering if you'd be interested in something like this for OMR.
All reactions