RDS-TCP: Do not bloat sndbuf/rcvbuf in rds_tcp_tune
authorSowmini Varadhan <sowmini.varadhan@oracle.com>
Wed, 30 Sep 2015 20:54:08 +0000 (16:54 -0400)
committerDavid S. Miller <davem@davemloft.net>
Mon, 5 Oct 2015 10:34:53 +0000 (03:34 -0700)
commit1edd6a14d24f21b8b478970c63a243a08e2b55b0
tree69b0d3d0619f70911f9dee2b98bbc9072fc14b50
parent3b20fc389705a4c959adebc494578cb99bb8be9e
RDS-TCP: Do not bloat sndbuf/rcvbuf in rds_tcp_tune

Using the value of RDS_TCP_DEFAULT_BUFSIZE (128K)
clobbers efficient use of TSO because it inflates the size_goal
that is computed in tcp_sendmsg/tcp_sendpage and skews packet
latency, and the default values for these parameters actually
results in significantly better performance.

In request-response tests using rds-stress with a packet size of
100K with 16 threads (test parameters -q 100000 -a 256 -t16 -d16)
between a single pair of IP addresses achieves a throughput of
6-8 Gbps. Without this patch, throughput maxes at 2-3 Gbps under
equivalent conditions on these platforms.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/rds/tcp.c