Intelligent RAID-0.

[RKumar]

<DIV class=postcolor>The problem with RAID-0's performance is that it heavily depends on the STRIPE size we choose.

Since STRIPE size is not configurable, once the array is built and data is stored, we will have to STICK with that STRIPE.

What i suggest is that instead of HARDWARE controller, can go with WinXP's Software RAID using Dynamic Disks.

Is it possible [may be using a custom Storage class driver], to have Intelligent STRIPE size
- either on a PER FILE basis
- or using a set of rules defined by user [considering file size, file type, ...]

This way we will always get best performance by dividing the "WORK" among "ALL" the RAID-0 array drives equally.

Did any one give this a thought yet? </DIV>

 

[WinkOsmosis]

Yes.

 

[RKumar]

Originally posted by: WinkOsmosis
Yes.

Could you elaborate please?

 

[Nebor]

Software RAID is slow.... that's why no one uses it.

 

[RKumar]

Originally posted by: Nebor
Software RAID is slow.... that's why no one uses it.

But definition of "Slow" is changed now i guess with new pentium "C" 800FSB HT enabled CPUs?

 

[Nebor]

But it's still slower than a hardware raid solution, which is free with a good motherboard.

 

[WobbleWobble]

Originally posted by: Nebor
But it's still slower than a hardware raid solution, which is free with a good motherboard.

I thought motherboard solutions were still software RAID solutions? It just allows you to boot to an OS in RAID. As far I know, it still utilizes much more CPU time than a "hardware" RAID controller.

 

[sharkeeper]

Too much overhead for potential gain. Efforts to optimise caching algorithms and firmware matching to master host is of higher priority. This is (of course) for real hardware solutions, not those found on desktop mainboards.

Cheers!