Veritas-bu

[Veritas-bu] 9940B

2006-10-14 03:38:55
Subject: [Veritas-bu] 9940B
From: ida3248b at post.cybercity.dk (ida3248b AT post.cybercity DOT dk)
Date: Sat, 14 Oct 2006 09:38:55 +0200
In my experience there is quite a performance improvement to be had on the 
STK9840/9940 drives, the problem usually is do the media servers has enough 
resources. 

Regards
Michael

On Fri, 13 Oct 2006 18:08:28 -0400, bob944 wrote
> > Was wondering whether anyone knows the max I/O buffer size for the STK
> > 9940B ?
> >  
> > Trying to fine tune some media servers using the SIZE_DATA_BUFFERS and
> > NUMBER_DATA_BUFFERS.
> >  
> > And was thinking the drive would obviously determine what the settings
> > should be.
> 
> That's unlikely to be correct.  How does the drive subdivide its buffer
> space (or does it?)?  How does it use it--wait for it to fill, then
> compute and write, waiting for it to empty before accepting more data
> (both unlikely)?  Would it wait to fill before looking to find out it
> contained 64KB of zeros?
> 
> The point is that without knowing the drive firmware, we have no idea
> how the drive uses its space--and if there were a performance advantage
> to be had by matching buffer sizes, the vendor would certainly tell you
> about it.  (BTW, the one modern drive firmware I have studied works
> nothing like a fixed buffer, but much more like a round-robin queue,
> optimized for continuous input of data and [de]compression at one "end"
> while writing out the other.  Its size is relevant only for avoiding
> over/under-run; it's rate that matters.)  Your objective is to never 
> let the drive run dry on write, and never let it fill on read.  Matching
> buffer sizes doesn't do either.
> 
> Buffer tuning results for serverA in infrastructureA are often quite
> different from best results for serverB/infrastructureB with the same
> tape drives.  If there's a better way to determine them than
> empirically, I'd like to know it.  The only correct answer IMO is to
> open the Tuning Guide and do the work.  After you do it the right 
> way, you have the numbers to understand your infrastructure and 
> predict the results of changes to it.
> 
> _______________________________________________
> Veritas-bu maillist  -  Veritas-bu at mailman.eng.auburn.edu
> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu


--
Cybercity Webhosting (http://www.cybercity.dk)


<Prev in Thread] Current Thread [Next in Thread>