From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <bef41a2498bb24f3a30ad88fd781f20f@plan9.bell-labs.com>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] Standalone cpu/auth..
From: rsc@plan9.bell-labs.com
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Date: Wed, 15 May 2002 21:21:33 -0400
Topicbox-Message-UUID: 93de991c-eaca-11e9-9e20-41e7f4b1d025

the drawterm sessions should be slower,
since they are going through srvold9p rather
than plugging the network connection
into the kernel directly.  it shouldn't be very
much slower, but perhaps a little.  if you're
comparing drawterm-linux->3e vs ->4e,
i don't think you should see a big difference.

if you're comparing drawterm-linux->3e with
drawterm-windows->4e, you will see a big difference.
repeating yesterday's theme, the 4e drawterm sources
from windows now disable nagle's algorithm,
which makes an enormous difference.

replace /sys/src/cmd/unix/drawterm/devip-unix.c:/^so_socket
with this one and recompile.  you should see a big
difference.

int
so_socket(int type)
{
	int fd, one;

	switch(type) {
	default:
		error("bad protocol type");
	case S_TCP:
		type = SOCK_STREAM;
		break;
	case S_UDP:
		type = SOCK_DGRAM;
		break;
	}

	fd = socket(AF_INET, type, 0);
	if(fd < 0)
		error(strerror(errno));

	one = 1;
	if(setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, (char*)&one, sizeof(one)) > 0)
		print("setsockopt: %s", sys_errlist[errno]);

	return fd;
}

russ