Re: misc/50046

[Frank Wille <frank%phoenix.owl.de@localhost>]

The following reply was made to PR misc/50046; it has been noted by GNATS.

From: Frank Wille <frank%phoenix.owl.de@localhost>
To: Christos Zoulas <christos%zoulas.com@localhost>,
	gnats-bugs%NetBSD.org@localhost,
	misc-bug-people%netbsd.org@localhost,
	gnats-admin%netbsd.org@localhost,
	netbsd-bugs%netbsd.org@localhost
Cc: 
Subject: Re: misc/50046
Date: Tue, 21 Jul 2015 17:01:47 +0200

 Christos Zoulas wrote:
 
 > Can you please test the stdbuf change I posted on tech-userlevel
 > (compile and re-install libc), together with adding:
 >
 > export STDBUF0=L
 > export STDBUF1=L
 > export STDBUF2=L
 >
 > in the beginning of rc and reverting rc and rc.subr to pre-pinger
 > changes?
 
 I did some test runs today on an Amiga 1200, 68030/40MHz with 64MB. The
 multiuser boot process takes here 8 minutes and 44 seconds, which was less
 than half the time with NetBSD 6.
 
 
 In the first tests I reverted /etc/rc to 1.167, but left the _rc_pid
 definition in line 87 in. Otherwise booting will hang (it is needed by
 rc.subr). Results ("new clib" includes your stdbuf changes):
 
 old clib: 7:00 min.
 new clib: 6:57 min.
 new clib with STDBUFn=L: 6:58 min.
 new clib with STDBUFn=U: 7:01 min.
 
 The stdbuf change doesn't make a big difference here. But removing the "nop"
 pinger code is already noticable.
 
 
 In the next tests I reverted /etc/rc to 1.167 and /etc/rc.subr to 1.93, as
 suggested by Arto Huusko:
 
 old clib: 3:44 min.
 new clib: 3:38 min.
 
 So I can confirm that reverting /etc/rc.subr is the key. Would be great to
 find a solution here.
 
 -- 
 Frank Wille