From 786bf56c01c6a8757bd9dbde670c3dbd600a600d Mon Sep 17 00:00:00 2001
From: sageweil <sageweil@29311d96-e01e-0410-9327-a35deaab8ce9>
Date: Wed, 21 Mar 2007 21:20:47 +0000
Subject: [PATCH] * ms_tcp_nodelay

git-svn-id: https://ceph.svn.sf.net/svnroot/ceph@1278 29311d96-e01e-0410-9327-a35deaab8ce9
---
 branches/aleung/security1/ceph/config.cc             |  1 +
 branches/aleung/security1/ceph/config.h              |  1 +
 .../aleung/security1/ceph/msg/SimpleMessenger.cc     | 12 +++++++++++-
 3 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/branches/aleung/security1/ceph/config.cc b/branches/aleung/security1/ceph/config.cc
index 4302b3f5c16..3f96ab6961b 100644
--- a/branches/aleung/security1/ceph/config.cc
+++ b/branches/aleung/security1/ceph/config.cc
@@ -118,6 +118,7 @@ md_config_t g_conf = {
   ms_overlay_clients: false,
 
   ms_die_on_failure: false,
+  ms_tcp_nodelay: true,
 
   /*tcp_skip_rank0: false,
   tcp_overlay_clients: false,  // over osds!
diff --git a/branches/aleung/security1/ceph/config.h b/branches/aleung/security1/ceph/config.h
index 701085eb61e..ce586d71d55 100644
--- a/branches/aleung/security1/ceph/config.h
+++ b/branches/aleung/security1/ceph/config.h
@@ -105,6 +105,7 @@ struct md_config_t {
   bool ms_skip_rank0;
   bool ms_overlay_clients;
   bool ms_die_on_failure;
+  bool ms_tcp_nodelay;
 
   // mon
   int mon_tick_interval;
diff --git a/branches/aleung/security1/ceph/msg/SimpleMessenger.cc b/branches/aleung/security1/ceph/msg/SimpleMessenger.cc
index 73632fc2f74..6f32b67afa3 100644
--- a/branches/aleung/security1/ceph/msg/SimpleMessenger.cc
+++ b/branches/aleung/security1/ceph/msg/SimpleMessenger.cc
@@ -18,6 +18,8 @@
 #include <sys/stat.h>
 #include <fcntl.h>
 #include <signal.h>
+#include <sys/socket.h>
+#include <netinet/tcp.h>
 
 #include "config.h"
 
@@ -413,6 +415,14 @@ void Rank::Pipe::writer()
     }
   }
 
+  // disable Nagle algorithm
+  if (g_conf.ms_tcp_nodelay) {
+    int flag = 1;
+    int r = ::setsockopt(sd, IPPROTO_TCP, TCP_NODELAY, (char*)&flag, sizeof(flag));
+    if (r < 0) 
+      dout(0) << "pipe(" << peer_addr << ' ' << this << ").writer couldn't set TCP_NODELAY: " << strerror(errno) << endl;
+  }
+
   // loop.
   lock.Lock();
   while (!q.empty() || !done) {
@@ -621,7 +631,7 @@ int Rank::Pipe::write_message(Message *m)
   }
 
 #else
-  if (0) {
+  if (1) {
     // one big chunk
     env->nchunks = -blist.length();