kernel_optimize_test/arch/alpha
Eric Dumazet 2c8c56e15d net: introduce SO_INCOMING_CPU
Alternative to RPS/RFS is to use hardware support for multiple
queues.

Then split a set of million of sockets into worker threads, each
one using epoll() to manage events on its own socket pool.

Ideally, we want one thread per RX/TX queue/cpu, but we have no way to
know after accept() or connect() on which queue/cpu a socket is managed.

We normally use one cpu per RX queue (IRQ smp_affinity being properly
set), so remembering on socket structure which cpu delivered last packet
is enough to solve the problem.

After accept(), connect(), or even file descriptor passing around
processes, applications can use :

 int cpu;
 socklen_t len = sizeof(cpu);

 getsockopt(fd, SOL_SOCKET, SO_INCOMING_CPU, &cpu, &len);

And use this information to put the socket into the right silo
for optimal performance, as all networking stack should run
on the appropriate cpu, without need to send IPI (RPS/RFS).

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-11 13:00:06 -05:00
..
boot alpha: boot: fix build breakage introduced by system.h disintegration 2013-03-08 15:05:34 -08:00
include net: introduce SO_INCOMING_CPU 2014-11-11 13:00:06 -05:00
kernel Merge git://git.infradead.org/users/eparis/audit 2014-10-19 16:25:56 -07:00
lib alpha: fix broken network checksum 2014-01-31 09:21:55 -08:00
math-emu
mm arch: mm: pass userspace fault flag to generic fault handler 2013-09-12 15:38:01 -07:00
oprofile oprofilefs_create_...() do not need superblock argument 2013-09-03 22:52:48 -04:00
defconfig
Kconfig Merge git://git.infradead.org/users/eparis/audit 2014-04-12 12:38:53 -07:00
Kconfig.debug
Makefile alpha: makefile: don't enforce small data model for kernel builds 2013-04-07 12:59:30 -07:00