Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 11 May 1999 19:47:08 +0400 (MSD)
From:      "Sergey Ayukov (mailing lists)" <asv1@crydee.sai.msu.ru>
To:        Doug Rabson <dfr@nlsystems.com>
Cc:        stable@FreeBSD.ORG
Subject:   Re: NFS question..
Message-ID:  <Pine.BSF.4.05.9905111941590.10169-100000@crydee.sai.msu.ru>
In-Reply-To: <Pine.BSF.4.05.9905102002340.447-100000@herring.nlsystems.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 10 May 1999, Doug Rabson wrote:

> > > It all depends on the value which you place on the data which the clients
> > > are writing.
> > 
> > Yes, sure. In ideal case, I would love to mount some directories with less
> > reliable settings and some with more reliable... This reminds me about
> > ages long discussion on whether writes should be cached at all (it was
> > thriving when DOS Smartdrive finally got an option to enable write
> > caching). My opinion is that they should be cached. After all, it is
> > impossible to get good performance out of NFSv2 when not doing write
> > caching. Whether you will rely on UPS or just pray for data to be safe is
> > another question.
> 
> The data is cached (subsequent reads will come from the cache). I don't
> expect FreeBSD's defaults to change on this since it does violate the spec
> and a much better (safer) fix is to use NFSv3.

There are cases when NFSv3 is simply not an option (can't upgrade
clients). IMHO, the filesystem which does fflush() after each write()
(effectively, this is a requirement for NFSv2) is seriously broken in
design.
 
> > > Could I have a copy of your test program? The 100Mbit performance ought to
> > > be better than this.
> > 
> > Test program is File Commander for OS/2, available from many places, e.g. 
> > ftp://hobbes.nmsu.edu/pub/incoming/fc2_210.zip
> 
> Hmm. Not too useful to me since I don't run OS/2. I wonder what sized
> requests are made by the OS/2 NFS client. That can affect performance
> considerably. Without knowing the pattern of NFS calls the test makes, it
> is impossible to speculate on what is causing the performance loss.

The problem has been solved! (see my other message). Answer:
sysctl -w vfs.nfs.gatherdelay=0

> > > I would have thought the OS/2 client could use SMB.  I thought the
> > > performance of samba was pretty good on FreeBSD. Perhaps it could be tuned
> > > a bit (samba has a boatload of tuning parameters).
> > 
> > I don't know why I am having such a bad luck with FreeBSD, but I am only
> > getting about 300KB/s on writes over 10MBit network while exchange between
> > Windoze machines yields about 900KB/s. Someday I will try SMB client on
> > OS/2, but I was pretty happy with NFS -- until I switched to FreeBSD.
> 
> Try reading the Samba documentation (in the docs/textdocs subdirectory of
> the samba source code).

Samba started to perform reasonably well (870KB/s on write via 10MBit/s)
after "sysctl -w net.inet.tcp.delayed_ack=0". I wonder why an option in
the smb.conf had no effect.

---------------------------------------------------------------------------
Dr. Sergey Ayukov                          Sternberg Astronomical Institute
http://www.ayukov.com                                        Moscow, Russia
http://crydee.sai.msu.ru/index-asv.html



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.05.9905111941590.10169-100000>