All else equal, RAID-5 is likely to be the biggest problem. RAID-5/6 are pretty 
bad for write performance, which will impact journal writes and large merges. 
CPF does quite a few journal writes, and merges just happen.

Here's a test you might try: back up your old forests to the new device, and 
then restore them to a set of newly-created forests on the same device. Force a 
merge on all three, and see how long it takes to finish. You can compare with 
the old devices, which will give you a good idea of just how much your new 
device configuration would impact performance. All things may not be equal, so 
it may be an improvement despite my knee-jerk reaction.

-- Mike

On 15 Jul 2012, at 11:51 , Tim Meagher wrote:

> Hi Michael,
> 
> Thanks for following up on this.  My database currently has approximately 8M
> documents taking up over 460BG of space split across 3 forests on 3 separate
> 500 GB devices.  I retain documents at various phases of production
> processed by more than 1 CPF domain (input, intermediate which takes a while
> to produce, and final formats).  The plan is to go to a single 1.5T local
> device in a RAID 5 configuration.  I was considering just copying the
> forests over to the new device to simplify the content migration, but my
> numbers don't work with your recommendations for performance.  Sounds like I
> should only have one or two destination forests, but without re-architecting
> my dataflow I way exceed the max number of GBs per database.  At what point
> does performance degradation become noticeable?
> 
> Thank you!
> 
> Tim Meagher
> 
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Michael
> Blakeley
> Sent: Sunday, July 15, 2012 2:15 PM
> To: MarkLogic Developer Discussion
> Subject: Re: [MarkLogic Dev General] Migrating content to another database
> 
> I know I am late with this reply, but I wanted to plug
> https://github.com/mblakele/task-rebalancer for this job. The downside is
> that it requires 5.0, or some patches for 4.2. But the 5.0 upgrade is very
> easy: if you are already planning on a move to 4.2, consider going to 5.0
> instead. You can defer reindexing until you are ready.
> 
> Assuming you can use it, the task-rebalancer should be faster than XQSync -
> as long as you allocate enough task server threads. The project also
> provides an example module to 'evacuate' a forest. You could modify it to
> evacuate your old forests, which would populate your new one(s).
> 
> Along those lines, I wouldn't let a single forest grow without bounds and
> multiple devices are good for performance. Generally speaking I try to make
> sure the database has:
> 
> * 1 forest per 2 CPU cores
> * no more than 200-GB each
> * no more than 32M documents each
> * no more than 2 forests per filesystem
> * no more than 1 forest per spindle
> 
> Some of these rules depend on the situation, too. With positions enabled,
> for example, I would try for something closer to 8M documents or 100-GB,
> whichever comes first. With RAID-1 or RAID-10 I would count drive-pairs as
> spindles. With RAID-5 or RAID-6 I would count RAID groups as spindles.
> 
> -- Mike
> 
> On 10 Jul 2012, at 19:15 , Tim Meagher wrote:
> 
>> Hi Folks,
>> 
>> I have over 150 GB of content in one database that is currently spread 
>> unevenly across 3 forests on 3 separate devices.  I need to migrate 
>> this content to a new database which uses one device with more than 
>> enough space for all of the content.  Since there is only one device, 
>> I'm wondering if there is any advantage or disadvantage to using 
>> multiple forests.  I think I should be able to simply copy the content 
>> by creating 3 forests in the new database and copying the forests 
>> over, but I'd like to know if this is not an optimal solution in which 
>> case I will need to be a little more resourceful about copying the content
> over.  Perhaps xqsync?
>> 
>> Tim Meagher
>> 
>> P.S. Using MOvign content from ML 4.1 to ML 4.2.  Sorry, not yet at 5!
>> 
>> _______________________________________________
>> General mailing list
>> [email protected]
>> http://developer.marklogic.com/mailman/listinfo/general
>> 
> 
> _______________________________________________
> General mailing list
> [email protected]
> http://developer.marklogic.com/mailman/listinfo/general
> 
> _______________________________________________
> General mailing list
> [email protected]
> http://developer.marklogic.com/mailman/listinfo/general
> 

_______________________________________________
General mailing list
[email protected]
http://developer.marklogic.com/mailman/listinfo/general

Reply via email to