>I hear ya on the benefits of SCSI over ATA/IDE.
>
>My concern is that this appears to be a fairly serious OS-level issue (hopefully MS is working on it?) As you pointed out, turning off lazy writes is not ideal from a performance standpoint, especially if doing so is not "sticky".
>
>With servers I'd be concerned that critical server-based apps might run into this issue as well. OTOH, if it only manifests itself on shutdown, and servers run 24/7, then it might never arise there, as the server *always* has enough time to eventually write the data to disk.
>

You've never dealt with some of my ABusers, the local power company, UI (Ultimately Infuriating), our RBOC, SNET (Screws Networks Every Time) or some of my software (Win2K is crashable even when it's not running my code, and it was installed and configured without my help. Trust me on this. If I did it, there's a good chance that I checked the HCL, and have enough RAM, a UPS, backup facilities and I usually sacrificed a PHM on the mousepad to appease Murphy...

>Modern OSs are supposed to issue an OS-level, system-wide "FLUSH" at shutdown; this *should* prevent these sorts of problems. It apppears that this is not the case for W2K and ATA/IDE. Unless...
>

ATA is not a wise choice for disk storage - it doesn't scale well, and is really suited to internal installation. I'm not surprised to see that it's being dinked to deliver performance - a lot of people were betting that ATA would be replaced by IEEE 1394 and USB, and when that fizzled, it got really important that the basic drive in typical systems perform well, so my unproven and very speculative feeling is that the ATAPI miniport tries to enable any features in the drive that might offer better performance with what users likely will have in place. If the big PC vendors who're putting Win2K Pro on their laptops and business desktop models end up performing poorly, guess what won't get sold?

>Question for you: In your testing, were you using machines that not only shut down, but powered off under OS control when shutdown was "complete"? These days, some ATA drives have buffers of 2MB or more, and I'm wondering if the drive is getting powered off before it can fully write its local buffer contents out to disk.

I don't think that's the problem.