🚀 go-pugleaf

On Thu, 4 Dec 1997 04:28:10, hahn@neurocog.lrdc.pitt.edu (Mark 
Hahn) wrote:

> : I'm not sure I understand you comment here, but if you're 
> : suggesting that IDE drives can do multiple requests, I disagree 
> : 100%.  SCSI drives can re-order requests to meet the physical 
> : limitations of the drive.  I've seen no evidence that IDE drives 
> : can do this.  I woulldn't expect it either.
> 
> there's limited opportunity for tagged command queueing to provide any
> benefit, since reasonable OSs already sort device request queues.  do you
> have some evidence that TCQ is a major win?  remember that ATA drives
> already do readahead and write-behind in nontrivial-sized caches.

If the drive (channel actually) has already been issued a request
all is halted until that request is served.  SCSI drives can 
order the responses to better match the requests. IDE simply 
cannot reorder requests.  If you want evidence, put your CDROM on
your primary IDE channel and look at your performance tank when 
it's in use.

Again, I'm not saying that SCSI is better than IDE.  In some 
circumstances it is.  On the desktop/client side (we're talking 
x86 here), I know of only one application that needs SCSI (DV). 
Certainly as x86 systems take over the UNIX world this will 
change.  On the server side, please, there is no competition.  
But, just because SCSI is better on a server does *not* mean it's
better on a client/stand-alone.  I don't want SCSI at home!

----
  Keith R. Williams
  krw@ibm.net