On 09/05/14 04:52, chaitanya lala wrote: > Hello, > > I was wondering if there has been any work/thoughts on trying to plumb NetBsd > TCP/IP rump stack into Linux containers ? What I mean by that is to make a > Linux sockets based application running inside LXC using the rump tcp/ip > stack as it's main stack, somehow bypassing the host Linux kernel's tcp/ip > stack and sending packets straight to the physical interface of the host. > This could be extremely helpful in a high performance setup as usually a > shared linux tcp/ip stack is a bottleneck (CONFIG_NET_NS is only a partial > remedy IMHO).
It's quite a surprising idea. At least to my knowledge you're the first to propose it. When I did the rumpuser-linuxkernel experiment, I didn't have any actually serious use cases in mind ;) > Assuming this can be done, what I cannot fathom is the way the LXC container > would bypass Linux TCP/IP. Any ideas or guidance is welcome. Is there any reason the data path couldn't work the same way as sockin [1] works? So you'd just provide alternative versions of "struct proto_ops", and select the rump kernel ones instead of the native Linux ones based on some flag. There are probably some details to handle for the config path, like interfaces and routing. Short getting hands dirty, it's difficult to guess how that will play out. [1] http://nxr.netbsd.org/source/xref/src/sys/rump/net/lib/libsockin/sockin.c br, antti ------------------------------------------------------------------------------ Is your legacy SCM system holding you back? Join Perforce May 7 to find out: • 3 signs your SCM is hindering your productivity • Requirements for releasing software faster • Expert tips and advice for migrating your SCM now http://p.sf.net/sfu/perforce _______________________________________________ rumpkernel-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/rumpkernel-users
