Subject: Re: TCP/IP tuning
To: None <tom@insolvencyhelp.org, current-users@netbsd.org>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: current-users
Date: 02/28/2004 15:57:00
On Sat, Feb 28, 2004 at 10:21:48AM -0500, Thomas Miller wrote:
> 
> ping -c 5 ftp.netbsd.org and 
> ping -c 5 -s 1400 ftp.netbsd.org 

I don't think using a 1400-byte packet is a particularly good way to
measure "delay", which is theoretically an instantaneous value.

> Sysctl says, however, 
> 
> # sysctl net.inet.tcp.recvspace
> net.inet.tcp.recvspace = 32768  
> 
> which, if that 32768 is bits,  seems to be less than 
> 1/10 of the BDP.  

It's bytes.

All that value is is the _default_ socket buffer size.  Applications
wanting more buffering can get it by using setsockopt, you can set it
system-wide with sysctl, or you can set it at compile-time for the
kernel.

Thor