Kniet nieder und betet ;) (cheetahs sind da :))

jo, suckt immer mehr, check this from google:

[quote]
From: Able <ablewang@able.per.sg>
Subject: Slow SCSI Performance
Date: Wed, 27 Nov 2002 04:59:40 +0800
Reply-To: protected
Message-ID: <n4o7uucboq0rf81oafb3t2909e81u83kfo@4ax.com>
X-Newsreader: Forte Agent 1.92/32.572
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Newsgroups: microsoft.public.windowsxp.general
NNTP-Posting-Host: ad202.166.7.93.magix.com.sg 202.166.7.93
Lines: 1


Does anyone knows when MS will release the fix for this problem?


Maximum PC - September issue

SCSI Slowdown

Your hard drive performance may be slower than you'd expect if you're
running Windows XP Professional or Home with a SCSI rig. That's what
users of high-priced and supposedly high-performance SCSI equipment
are apparently finding as they convert to Windows XP Pro or Home from
Windows 2000 or 98. In fact, Maximum PCs own crew ran into the XP/SCSI
bug this month when they first laid down the configuration for the
Dream Machine (see page 24). Instead of the nice-but-not-so-exotic ATA
RAID configuration that ended up in this year's dream rig, the
original configuration included a new hot-to-trot Seagate Cheetah
15K.3 drive. Packing 73.4GB of storage capacity 3.6ms seek times, and
a 15,000 spindle speed, the Cheetah may well be the fastest hard drive
around end can even motor past an ATA RAID configuration without
breaking a sweat. Indeed, using the quick-and-dirty ATTO hard drive
benchmark under Windows 2000 Professional, the Cheetah 15K.3 can't be
beat by anything in read or write speeds. Run the drive with Windows
XP Professional however, and the Cheetah 15K.3 is outperformed by a
5400rpm budget drive in write speeds of small files. The problem isn't
Seagate's new wonder drive - it's an incompatibility between Windows
XP Professional and certain disk intensive applications.

Only at hard drive fetish web sites such as Storagereview.com has the
problem begun boiling over as forum members search for solutions. A
previous NTFS fix doesn't solve the problem, nor does the upcoming
Service Pack 1. But don't start selling your equipment on eBay, just
yet. The Dog has learned from Microsoft that a fix is on the way.

Jeff Goldner, a development manager for the Windows-based group, told
The Dog that a combination of long-term goofs led to the SCSI
slowdown.

"We share some blame here. Goldner said "[with previous OS versions],
we told developer to set FILE_FLAG WRITE_THROUGH and FILE
FLAG_NO_BUFFERING to obtain better performance." Unfortunately, these
instructions don't jibe with Windows XP.

Goldner said the problem actually existed under Windows 2000 too, but
many users probably wouldn't have noticed it without turning off write
caching. Under Windows XP, the problem manifests because it's assumed
by Microsoft that the write through setting is not being used by the
application. But because of previous encouragement by Microsoft, many
applications that stress the hard drive (such as DV, AV, and benchmark
tests) request the flag nonetheless.

"We've dune some tests and have seen up to 10x performance hits
depending on the packet size, Goldner said. "It depends on the drives
and how they implement the cache." Unfortunately, the flag cannot
simply be removed from XP because its absence can affect file
integrity. Much of the problem can be addressed by recompiling the
applications with the flag turned oft or by using XP's compatibility
update feature. The updates can fool the applications into thinking
the flag is on, while actually switching it off. Goldner said
Microsoft expects to release updates within the next three months and
will also issue a more detailed Knowledge Base item. Woof.
[/quote]

removed

wie siehts aus mit dem mylex problem?
gibts dafür schon eine lösung?

Im Vergleich zu einer CPU sind solche Platten sehr genügsam (so um die 10 Watt, beim Hochfahren halt mehr).

Zitat von that

Im Vergleich zu einer CPU sind solche Platten sehr genügsam (so um die 10 Watt, beim Hochfahren halt mehr).

naja, 10W gelten für den idle zustand... wenn ein zugriff ist, dann zieht die platte natürlich mehr strom.

theoretisch kann der onboard controller schon schneller sein, z.b. so wies bei der neuen MCP-S southbridge für den nf2 chipsatz funtionieren wird: SATA direkt in der sb integriert, daher anbindung nicht als PCI gerät sondern direkt über HyperTransport

sollte ja mit einem SCSI auch funktionieren, aber ich glaub sowas gibts (noch?) nicht zu kaufen

Prinzipiell habt ihr recht, WENN der Controller in die SB integriert wäre, dann wäre er direkt über Hypertransport, V-Link, Mutiol etc., und nicht über den lahmen PCI Bus angebunden. Beim Intels kommender ICH5 SB (Canterwood SB) wird ein SATA-RAID Controller integriert sein und damit nicht mehr der Bandbreitenlimitierung unterliegen.
Nur ich bin jetzt erstens von der derzeitigen Situation ausgegangen, und zweitens von SCSI.

Und WENN es Onboard-Controller gibt, die über PCI-X angebunden sind, dann nur auf sauteuren INTEL-Serverboards, unter 600€ brauchst da gar nicht zu suchen anfangen.