aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorOlga Kornievskaia <aglo@citi.umich.edu>2008-10-21 14:13:47 -0400
committerJ. Bruce Fields <bfields@citi.umich.edu>2009-03-18 17:46:59 -0400
commit47a14ef1af48c696b214ac168f056ddc79793d0e (patch)
treef2f8fc4ae9b8cc142e3d4c7b1978d3128cce8bdd
parent026722c25e6eb018eab8b9a3c198c258f5b7a2e7 (diff)
svcrpc: take advantage of tcp autotuning
Allow the NFSv4 server to make use of TCP autotuning behaviour, which was previously disabled by setting the sk_userlocks variable. Set the receive buffers to be big enough to receive the whole RPC request, and set this for the listening socket, not the accept socket. Remove the code that readjusts the receive/send buffer sizes for the accepted socket. Previously this code was used to influence the TCP window management behaviour, which is no longer needed when autotuning is enabled. This can improve IO bandwidth on networks with high bandwidth-delay products, where a large tcp window is required. It also simplifies performance tuning, since getting adequate tcp buffers previously required increasing the number of nfsd threads. Signed-off-by: Olga Kornievskaia <aglo@citi.umich.edu> Cc: Jim Rees <rees@umich.edu> Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
-rw-r--r--net/sunrpc/svcsock.c35
1 files changed, 7 insertions, 28 deletions
diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index 5763e6460fea..7a2a90fb2e06 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -345,7 +345,6 @@ static void svc_sock_setbufsize(struct socket *sock, unsigned int snd,
345 lock_sock(sock->sk); 345 lock_sock(sock->sk);
346 sock->sk->sk_sndbuf = snd * 2; 346 sock->sk->sk_sndbuf = snd * 2;
347 sock->sk->sk_rcvbuf = rcv * 2; 347 sock->sk->sk_rcvbuf = rcv * 2;
348 sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK;
349 release_sock(sock->sk); 348 release_sock(sock->sk);
350#endif 349#endif
351} 350}
@@ -797,23 +796,6 @@ static int svc_tcp_recvfrom(struct svc_rqst *rqstp)
797 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), 796 test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags),
798 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); 797 test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags));
799 798
800 if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags))
801 /* sndbuf needs to have room for one request
802 * per thread, otherwise we can stall even when the
803 * network isn't a bottleneck.
804 *
805 * We count all threads rather than threads in a
806 * particular pool, which provides an upper bound
807 * on the number of threads which will access the socket.
808 *
809 * rcvbuf just needs to be able to hold a few requests.
810 * Normally they will be removed from the queue
811 * as soon a a complete request arrives.
812 */
813 svc_sock_setbufsize(svsk->sk_sock,
814 (serv->sv_nrthreads+3) * serv->sv_max_mesg,
815 3 * serv->sv_max_mesg);
816
817 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 799 clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
818 800
819 /* Receive data. If we haven't got the record length yet, get 801 /* Receive data. If we haven't got the record length yet, get
@@ -1061,15 +1043,6 @@ static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv)
1061 1043
1062 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; 1044 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1063 1045
1064 /* initialise setting must have enough space to
1065 * receive and respond to one request.
1066 * svc_tcp_recvfrom will re-adjust if necessary
1067 */
1068 svc_sock_setbufsize(svsk->sk_sock,
1069 3 * svsk->sk_xprt.xpt_server->sv_max_mesg,
1070 3 * svsk->sk_xprt.xpt_server->sv_max_mesg);
1071
1072 set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags);
1073 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); 1046 set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
1074 if (sk->sk_state != TCP_ESTABLISHED) 1047 if (sk->sk_state != TCP_ESTABLISHED)
1075 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); 1048 set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags);
@@ -1140,8 +1113,14 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
1140 /* Initialize the socket */ 1113 /* Initialize the socket */
1141 if (sock->type == SOCK_DGRAM) 1114 if (sock->type == SOCK_DGRAM)
1142 svc_udp_init(svsk, serv); 1115 svc_udp_init(svsk, serv);
1143 else 1116 else {
1117 /* initialise setting must have enough space to
1118 * receive and respond to one request.
1119 */
1120 svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg,
1121 4 * serv->sv_max_mesg);
1144 svc_tcp_init(svsk, serv); 1122 svc_tcp_init(svsk, serv);
1123 }
1145 1124
1146 /* 1125 /*
1147 * We start one listener per sv_serv. We want AF_INET 1126 * We start one listener per sv_serv. We want AF_INET