Nar Ganapathy - Windows, the IO Manager and Driver Model , Part I

Very cool.  Nar talked a bit about this last year at WinHEC.  Hope we get more details at this year's WinHEC next week.

My only concern about his performance theory that latency is ok as long as the bandwidth is still maximized is input devices.  With a Tablet for example the bandwidth requirement is low but latency (or lack of) is important.

I was doing some research into tradeoffs between hard drive SATA and SCSI technologies when I came across the fact that Server 2003 supported Raid 5 Striping something that is new I believe in the server product.

I was also looking at Tag Command Queing (TCQ) and SATA related NCQ for hard drives.

Does Server 2003 support TCQ or something akin to it during  management of Raid 5 or does the kernel leave all of this to disk controller hardware?

I ask the question because a number of motherboards manufacturers Ausus for one have implemented Raid 5 but have ignored TCQ I guess because the intended market is workstation. However Abit have implemented NCQ without featuring raid on a server board.

Should the kernel glue these two together somehow?

Another reason for asking this is one of vision - perhaps in the future distributed computing may become an everyday occasion. In this scenario workstation hard drive access begins to mimic that of a server in that jobs may be running from many diverse sources giving TCQ a role to play.

Try to ignore the urms