SUMMARY: a1000 vs d1000

From: Matthew Hannigan <mlh_at_zip.com.au>
Date: Wed Sep 11 2002 - 01:13:26 EDT
Thanks to
	Graham Wood 
	Bill McCaffrey 
	Darren Dunham
	Jay Lessert
	Jason Watson 


Short Summary: 
> 1.  Will the memory cache make the performance
> much better?  Considering that there is a maximum
> of 40Mb to cache 430+ GB!

A: It depends :-)  See the comments below.

> 2. Can you even use the memory cache on the a1000
> if you're not using the RAID features?

A: Yes (literally no, but really yes).  Since you
cannot use the A1000 without using the raid features,
you always get the benefit of the cache.  "Not using" the
raid features in this case would mean making each drive
a single concat.  Or I could make one or two giant raid0's.


Longer Summary:

Two responses leaned towards the D1000, one neutral
and two leaned towards the A1000.

The D1000 has two external scsi connections which
could make it more flexible and perhaps make it
perform better.

But as things turn out, we have the extra money for
the a1000 -- it was just within our budget, so it
looks like we'll be going with that.


I have further commments below amongst the replies below.




Graham Wood 


	[ re performance ]
	This all depends on how evenly spread over the 430GB the queries are.  If 
	you're looking at a database, then it is possible that the indexes will be 
	less than this size, and therefore be cached quite effectively.  The biggest 
	problem with any form of caching is that you are unable to accurately predict 
	anything.

	[ re memory cache ]
	No.  But since you can't use an A1000 without using the raid side of things, 
	its a totally irrelevant question.  The A1000 has a SCSI external connection, 
	but the disks are internally on 2 different SCSI busses, the same as the 
	D1000.


	For performance reasons, I'd recommend the D1000.  Instead of a single SCSI 
	connection to the 12 disks, you can have 2.  Also the A1000 would add an extra 
	layer of logic on top that you are not after.  The main advantage of the A1000 
	is that since the raid controller is hardware, you get very little performance 
	hit compared to software.  In your case you are not looking at raid, therefore 
	this is not relevant.  The sustainable transfer rate through a HDD these days 
	is a large fraction of the 40MB/s that is transferable through the SCSI 
	interface on the back of an A/D1000.  This means that from a purely data 
	throughput point of view (quite likely to be the bottleneck with a good server 
	and a datawarehousing system) you are best maximising your connectivity to the 
	disks.  

	Also you have to remember that your RDBM will be doing caching with all the 
	available system ram that it can too.  You are probably best going for the 
	D1000 and putting any money saved into extra money for the server.  

	Final thought - if you really need top performance from the system, I would 
	consider going for a T3 instead of the A/D1000.  The fibre connection is more 
	than double the speed, and there is an easy upgrade path to mirroring/etc if 
	you want to do that later.

me: T3 is way out of our budget for this project.



Bill McCaffrey 

	We have found that the D1000 out-performs the A1000 with an oracle database.
	We have not used then without mirroring, but the D1000 with software raid
	had better throughput than the A1000 with hardware raid.




Darren Dunham 

	You can't even afford to dedicate 1 of the 12 drives to parity?  That
	gives you ~396GB of redundant data on the A1000.

me: that may be possible but I would think that the one parity drive
would be a performance bottleneck.  Performance is more important
than uptime for this project.  The data can be just reimported from
our production system in case of disaster.


	[ re cache and performance ]

	It should make writes perform better, but probably not reads to any
	significant amount.  The more RAM you have in the host, the more that is
	true.  



Jay Lessert 

	I've used an A1000 in "one-big-RAID5" mode.  If you're doing 5-10%
	writes, the NV cache makes a huge difference, and the A1000 really
	rocks.

	10% writes is a pretty big percentage, by the way.




Jason Watson 


	[ re performance ]

	Unless you are doing heavy data throughput (such as video editing or 
	completely random) reads and 
	writes the cache will help.  The cache needs only to buffer the activity 
	to and from the disk; it doesn't have to be huge to accomodate the 
	size of the array.  For 
	your purposes I think 40 MB would do well.

	[ re use of cache ]

	I don't see any reason why not.  You would be putting the array into 
	spanning or RAID-0.  If the RAID controller is worth its salt it will 
	cache RAID-0 just as well as any of the other RAID levels.

	The other advantage the A1000 will have is that it off-loads the RAID-0 
	work from the host CPU to the RAID controller.  That can make a noticable 
	difference between the two.  Especially if the attached box is really 
	hitting its processor for the data mining.

--
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Wed Sep 11 01:16:50 2002

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:42:54 EST