Subject: Re: finally i've installed NetBSD-1.5
To: Manuel Bouyer <firstname.lastname@example.org>
From: Wojciech Puchar <email@example.com>
Date: 01/21/2001 14:48:29
> > have leased line with 33600 modem, chatscript and pppd config already
> > works in linux, and i found netbsd pppd almost identical to linux)
> Yes, I think you can put it there, but I never tried. I have installed
> mgetty and have it start pppd when needed (mgetty in started via /etc/ttys).
> Anoter option would be to start pppd from the startup script (in think
> setting ppp_peers in /etc/rc.conf to a valid /etc/ppp/peers/foo file
> is enouth), with 'nodetach' changed to 'persist'. Then pppd will go in
> background and restart the connection when it's terminated.
> > 2) why it is so fast :)?
> > especially - disk transfer from UDMA/66 disk (dd if=/dev/wd1d of=/dev/null
> > bs=64k count=5000 give me 26MB/s, and 23MB/s while doing 2 cp processes
> > on first disk)
> > i have (now) STANDARD generic kernel, while in linux i have -mpentiumpro
> > compiled kernel with latest ide-patches etc.. and i'm getting 21MB/s on
> > unloaded system maximum.
> Don't know, I didn't look closely at the linux driver :)
i think it's due to raw device - in linux you can't read without buffering
and lot of memcpy operation.
> > quite responsible even of heavy disk load including writes.
> The NetBSD VM is better than linux's under load (on the other hand, it's a bit
> slower when you have plenty of RAM - need to chose :). Also, linux has
> a dynamically sized buffer cache which allows to flush pages to RAM to
> use them as buffer cache when there's a lot of filesystem activity.
> So under linux pages of the program you're using can be flushed to swap to
> be remplaced by buffer caches.
i know it - especially after copying for eg. whole CD-ROM to disk i have a
lot of swapped programs while having plenty of RAM.
> > is it possible to easy recompile it all with -mpentiumpro -O2
> > (excluding pkgsrc for now) and make resulting gcc defaults to -mpentiumpro
> > (as i have in linux)
> Yes, I think you can set 'DBG= -mpentiumpro -O2' in /etc/mk.conf. You may
> also need 'HOST_CFLAGS= -mpentiumpro -O2'.
> > 5) for what is all this strange option for ffs like cylinder size, groups
> > etc. etc. etc.
> > for at least 10 years there's no drive with real geometry. how to turn off
> > this "optimization" in ffs or it does not matter?
> Who told you we don't use 10 years disks ? :) But I don't think it does
nobody. and it's very nice netbsd is able to efficiently make use of them.
> matter much. I usually change cpg/sgs from 16 to 256 in the disklabel
> (don't know if there's a way to change this from sysinst) to have
i know now how to install netbsd almost without sysinst :)
no problem - but at next reinstallation :)
> larger cylinder groups but that's all.
> > why there is so many seeks when cp large files making maximum 3MB/s
> > copying speed. can it somewhere be set to delay writes little more to get
> > longer writes?
> Linux uses async mounts by default (anything is cached and written later).
> By default NetBSD uses aysnc for data and sync for metadata. This ensures that
> the filesystem on disk is always in a coherent state (you can hit 'reset' even
> under heavy write, your system will always reboot without troubles).
> NetBSD can also mount filesystems async, but this is not recommended because
> you can trash your filesystem in case of unclean shutdown at the wrong time.
> But you can use soft updates (mount the filesystem with option 'softdep',
> change 'rw' to 'rw,softdep' in the fstab): writes are delayed but are still
> ordered, which keep things on disk in a coherent state. Aysnc performances,
> but with reliability :)
it was tested with softdep enabled!
i think it's a problem netbsd do maximum of something like 64-128kB writes
not more. can it be changed?
> > with about 80-90% filled disk (so cleanerd would be activated some time)
> > can lfs be used on root partition
> I would't use the LFS in 1.5 for anything serious. In -current it's a bit
> better but it's still experimental. FFS with soft updates should give you
> 'good enouth' performances :)
yes it is faster than ext2. the only slower thing is large file copy..
> > b: needed swap space
> > h: /boot space where kernel will be placed (max 10MB)
> > as i had in linux?
> On NetBSD, kernel and boot loader are in '/', not 'boot'.
> In /bin and /sbin you've some statically linked binaries that can help a lot
> in case of disaster.
> What I would do is a 50-100MB partition for /, and a separate partition
> for /usr with most space (then you can have /usr/home).
and can be /usr/var?
/var (at least in linux) can be VERY large because of /var/spool/mail,