Hi All, Is anyone making a similar effort? Why don't we work together and share the status?
I have been looking at memory mapped netlink introduced in Linux 3.10 and trying to understand its advantages and disadvantages examining stability, performance, programmability, etc. Since I'm focusing on the OVS kernel module and the netlink communication right now, I'm not using the OVS user-space utilities but another simple library to handle the netlink protocols. Even though it's still work in progress, all the materials are available at github, A patch for the OVS kernel module to use memory mapped netlink, https://github.com/takayuki/openvswitch/tree/nl_mmap And, a simple library to handle OVS netlink protocols, A Minimalistic Approach to Open vSwitch Datapath https://github.com/takayuki/libdpif By the way, I was quickly skimming the archive of the list, when I found another effort to use memory mapped netlink with OVS, [ovs-discuss] Any plan to support memory mapped netlink? http://openvswitch.org/pipermail/discuss/2013-July/010759.html The patch attached to the message contains only updates in user-space, lib/netlink-protocol.h and lib/netlink-socket.c. But, I think some updates will be needed in kernel-space to take fully advantage of memory mapped I/O in netlink, in particular, in communication from kernel to user space. What do you think? Takayuki Usui Midokura _______________________________________________ discuss mailing list [email protected] http://openvswitch.org/mailman/listinfo/discuss
