Subject: Re: Request for Comments on article
To: Andrew Gillham <firstname.lastname@example.org>
From: Manuel Bouyer <email@example.com>
Date: 02/27/2000 21:21:54
On Sat, Feb 26, 2000 at 12:04:08PM -0500, Andrew Gillham wrote:
> Right, I think this section needs clarification. It needs to be crystal
> clear that the NetBSD issue is with Adaptec 7890 type cards, not U2W or
> LVD in general. Currently the Adaptec 2940UW is probably the fasted controller NetBSD supports and it is capable of 40MB/s not the piddly 10MB/s suggested by
> "narrow scsi" mentioned in the article.
> AFAIK the ISP based boards support U2W/LVD under NetBSD 1.4.2, and potentially
> something based on the NCR/Symbios/LSI chipset. (53c875?)
The 875 is supported but is only UW. The NCR U2W is the 895. I tested a
-current boot floppy on a machine (HP NetServer something) with 2 895, it
seemed to work fine, although I had only UW disks to test. For the story,
linux had troubles using one of the 2 SCSI bus on this machine (didn't probe
the disk) where NetBSD used both without troubles.
> Good point. I also think that mentioning the BSD NFS code, particularly
> when compared to Linux in a lab environment where all machines are probably
> mounting big chunks of their filesystems from a central server.
Because of users pressure my PC client are linux machines, and we have small
annoyances with NFS: poor write performances, coherency problems.
Because of Linux's agressive caching you can create or modify a file on a
client, the changes will show up on the server only a few seconds later,
it's annoying if you run your editor on the local machine and compile on
a remote host. More annoying is that, because of this cache implementation,
file creation over NFS is not atomic, this break the 'dot locking' protocol
(especially, don't read your mail from a linux machine !!!).
Manuel Bouyer, LIP6, Universite Paris VI. Manuel.Bouyer@lip6.fr