Re: [PATCH openvswitch v3] netlink: Implement & enable memory mapped netlink i/o

[Ben Pfaff <blp@nicira.com>]

On Tue, Dec 03, 2013 at 12:19:02PM +0100, Thomas Graf wrote:
> Based on the initial patch by Cong Wang posted a couple of months
> ago.
> 
> This is the user space counterpart needed for the kernel patch
> '[PATCH net-next 3/8] openvswitch: Enable memory mapped Netlink i/o'
> 
> Allows the kernel to construct Netlink messages on memory mapped
> buffers and thus avoids copying. The functionality is enabled on
> sockets used for unicast traffic.
> 
> Further optimizations are possible by avoiding the copy into the
> ofpbuf after reading.
> 
> Signed-off-by: Thomas Graf <tgraf@redhat.com>

If I'm doing the calculations correctly, this mmaps 8 MB per ring-based
Netlink socket on a system with 4 kB pages.  OVS currently creates one
Netlink socket for each datapath port.  With 1000 ports (a moderate
number; we sometimes test with more), that is 8 GB of address space.  On
a 32-bit architecture that is impossible.  On a 64-bit architecture it
is possible but it may reserve an actual 8 GB of RAM: OVS often runs
with mlockall() since it is something of a soft real-time system (users
don't want their packet delivery delayed to page data back in).

Do you have any thoughts about this issue?