Re: Wanted: RAID Benchmarks

In article <3dtbb4$p08_at_deep.rsoft.bc.ca> a07893_at_giant.rsoft.bc.ca (Tim Bray) writes:
>Gary M. Watson writes:
>
>>The RAID Advisory Board (HP is a member) has commissioned Peer Protocols
>>to write a benchmarking suite for RAID arrays, as well as a test suite.
>
>Uh, I don't get it. A RAID is a disk, right, only faster and more robust,
>right? Do I care what it is inside? There are a variety of pretty good
>I/O benchmarks out there, including <blush> my own Bonnie, IOZONE, and the
>big one from Patterson et al that everyone says is the bee's knees. These
>things all try to measure real-world things like the throughput of Unix
>filesystems. Which kinda what I think you'd run on a RAID, most times.
>Unless of course you're doing private-partition work like Oracle/Sybase
>etc., in which case you'd better try it out with Oracle/Sybase/etc.

I think the idea was that since RAID arrays do great on some things (much better than a single spindle) and crappy on other things (worse than a single spindle), anyone using a synthetic benchmark that took some polynomial combination of the things that work great and the things that don't wouln't tell the whole story. So, I think they planned to to split the benchmark up so the individual components could be weighed individually. This is helpful in determining what the impact of cache is as well as allowing more direct comparison of one brand of RAID vs. another brand of RAID.

>Or is this RAB stuff just a marketing exercise?

That's how I consider benchmarks in general! Remember:

• Gary's Inviolate Axiom of Computing #16: *
• *
• All synthetic benchmarks that compare non- *
• identical systems are worthless. *

  ---

But yes, I think if it were me, I'd just run a test of whatever application I use frequently, on some large test case, and look at those results.

I think, though, when one talks about "RAID Benchmarks" one should remember why one buys RAID -- system uptime, ease of maintenance, and data integrity. These are hard to put into a number that means much. But you need to look for things like hot swap, spare parts availability and economy, service, support, etc. Does the system have to be shut down to replace a fan? Can a network operator change a disk drive or do I have to call 1-800-MEGABUCKS to get a tech on site after hours? Are the power supplies redundant? What does a new one cost? What will it cost 5 years from now? How nasty is the software installation? What happens next year when I change from Novell to Windows NT? (Or from TRS-DOS to OS/VS?) (Or Unix to Macintosh? or whatever).

These issues are one of the few remaining reasons why shows like Comdex are worthwhile -- you can go from vendor to vendor and kick the tires of the various products. It's helpful to know what questions to ask, of course. I've written something on this topic and will post it here if anybody cares. It's only slightly slanted towards our products, IMO, and other vendors are welcome to pipe in and slant it some other which-way...

-- 
Gary M. Watson                              
Electronic Engineer                         Internet: trimm_at_netcom.com
Trimm Technologies                          ftp: netcom.com, pub/trimm 
Las Vegas, Nevada                           Phone: (800) 423-2024 x2115 
** Support/Sales for Trimm Enclosures: trimmhlp_at_netcom.com (Dick Cavanaugh) **