On Thu, 13 Apr 2006 15:38:44 +0100
Dieter <[EMAIL PROTECTED]> wrote:

> > > > > I just got 49194139 bytes/sec reading a large file (~10GB) on FreeBSD.
> > > > > Linux on the same machine could only manage 35741200 bytes/sec.
> > > >=20
> > > > Which controller, kernel, and file system are you using ?
> > >=20
> > > nforce4-ultra Seagate 7200rpm SATA
> > > 6.0 FFS aka UFS
> > 
> > UFS on Linux ?
> > I'm pretty sure your benchmark only shows that Linux' implementation of UFS
> > is weaker than FreeBSD's, which is not surprising.
> 
> Linux: 2.6.16 and the default penguinix filesystem

I'll take a guess and say ext3. 

> FreeBSD read the exact same file from Linux's fs at 37194917 bytes/sec.
> So FreeBSD can read files off a Linux fs faster than Linux can.

Very weird. This may also depend on the I/O scheduler, which one are you using ?

> Linux refuses to mount the UFS fs.

Not too surprising, since it's FFS 6.0.

> This wasn't intended to be a "BSD is better than Linux" thing.

I rather hope so.

> > > > > The biggest disk speed problem today for SATA is the lack of NCQ supp=
> > ort.
> > > > > Developers are having trouble getting info on how to access NCQ for
> > > > > various controllers.  Thus, this is an area where an open-source SATA
> > > > > controller could help significantly.
> > > >=20
> > > > This market is taken or will be in the near future, at least at the
> > > > low-end :
> > >=20
> > > low end vaporware?
> > 
> > SiliconImage chipsets are not vaporware. Also, what is in the pipeline now
> > should be available for sale a few months from now, so I would not count
> > AHCI as vaporware either.
> 
> "or will be in the near future" sounded like vaporware.
> 
> Has SiliconImage gotten their act together?  Some of their SATA chips
> are buggy.  Luckily the one I have is merely very slow.  (2 drives
> together total 40 MB/s dd-ing sequential sectors from the raw device, no
> buffer cache, no filesystem.  Compared to the nf4u which does 65-70MB
> *per drive* running 4 drives at once.  Same make&model drives) 

What I heard was that SiliconImage controllers on some NF4 mainboards
behave badly. The benchmarks I've seen on good SiliconImage boards are
surprising, with Sil-3112 and 3114 being faster than anything when they
got out, including even Intel ICH6. 

> > Antivirus servers running on F/OSS prove that F/OSS is superior in
> > terms of security than proprietary software from Redmond. I do not
> > see how this helps WHG III.
> 
> If you help the morons keep the virus away from the virus servers,
> they will continue to run virus server.

Or buy existing proprietary antivirus technology to keep the virus at
bay. F/OSS loses a foothold.

> > People started
> > using Windows when F/OSS was not an option.
> 
> Unix has been around far longer than virus server.  Unix has never,
> even in the early days, had the level of problems that virus server
> has.

UNIX was not as affordable as Win95/NT 10 years ago, by a wide margin.
People got hooked on the MS ersatz. No need to rewrite history, this
is what happened and we all wish it had not.

> > Whole companies cannot
> > make the switch from proprietary (especially Windows, not UNIX)
> > to F/OSS
> 
> Proprietary UNIX is not a problem.  Having source is better, certainly.
> but proprietary UNIX doesn't create massive amounts of problems,
> the companies that sell it aren't constantly being hauled into court
> for illegal business practices, etc. etc. etc.

Yes, hence the "not UNIX" comment. Also, it's far easier to migrate
from UNIX to FOSS than Windows to FOSS. 

> > Let's help them and
> > make money at the same time.
> 
> Help them migrate away from virus server, yes.
> Help them keep virus server limping along, no.

I'm not sure willingly losing a market this way is a good idea. On the
other hand, even if antivirus software is typically very CPU-intensive,
I am not even sure there is a market in the first place :)

François
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to