Re: kern/45352

[Darren Reed <darrenr%fastmail.net@localhost>]

The following reply was made to PR kern/45352; it has been noted by GNATS.

From: Darren Reed <darrenr%fastmail.net@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: kern/45352
Date: Thu, 22 Sep 2011 09:43:30 +0200

 I've recompiled a kernel with a tty buffers set at 8192 and TTYHOG at 8192.
 
 It seems to work... try 8192 with the sysctl patches..
 
 The "why" would appear to be the way TTYHOG influences the tty being put 
 into blocking mode.
 
 A quick glance suggested that "TTYHOG/5" as being a magic number/chunk 
 size, although there are probably others too.
 
 Rather than mess around with different values for TTYHOG, it might be 
 worthwhile implementing a new "tty mode", activated by an ioctl, that 
 imposes different constraints. This would then be used primarily by pppd.
 
 
 I'm not convinced that even 8192 is "right" and that maybe larger 
 numbers are required for optimal performance with the current tty logic.
 
 What have FreeBSD and others done for this?