Re: [PATCH net-next v1] net: use a per task frag allocator

From: Eric Dumazet <redacted>
Date: Wed, 19 Sep 2012 18:56:01 +0200

From: Eric Dumazet <edumazet@google.com>

We currently use a per socket page reserve for tcp_sendmsg() operations.

This page is used to build fragments for skbs.

Its done to increase probability of coalescing small write() into
single segments in skbs still in write queue (not yet sent)

But it wastes a lot of memory for applications handling many mostly
idle sockets, since each socket holds one page in sk->sk_sndmsg_page

Its also quite inefficient to build TSO packets of 64KB, because we need
about 16 pages per skb on arches where PAGE_SIZE = 4096, so we hit
page allocator more than wanted.

This patch switches this frag allocator from socket to task structure,
and uses bigger pages.

(up to 32768 bytes per frag, thats order-3 pages on x86)

This increases TCP stream performance by 20% on loopback device,
but also benefits on other network devices, since 8x less frags are
mapped on transmit and unmapped on tx completion.

Its possible some TSO enabled hardware cant cope with bigger fragments,
but their ndo_start_xmit() should already handle this, splitting a
fragment in sub fragments, since some arches have PAGE_SIZE=65536

Successfully tested on various ethernet devices.
(ixgbe, igb, bnx2x, tg3, mellanox mlx4)

Followup patches can use this infrastructure in two other spots
and get rid of the socket sk_sndmsg_page.

Open for discussion : Should we fallback to smaller pages
if order-3 page allocations fail ?

Signed-off-by: Eric Dumazet <edumazet@google.com>

I like this a lot and I look forward to your upcoming changes to
convert the other two sk_sndmsg_page users as well, but I can't
apply this to net-next just yet.

The question on fallback is a good one and something we have
to resolve before applying this.

Note in particular that sk_allocation can be set to just about
anything, and this also has potential interaction issues with
SOCK_MEMALLOC.