Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

876

May 29th, 2008 06:00

diskeeper

May Diskeeper helps to improve the performance on a Windows 2003 Server. The Server has access to a DMX-3 Storagesystem. EMC has no Tools for defragment Disks for Symmetrix Systems.

385 Posts

May 30th, 2008 04:00

Don't confuse access methodology (cacheing) with file system structure. In short - the Symmetrix has NO impact on filesystem layout. It is really just providing a LUN to your host and servicing reads/writes using cache to increase the response time where possible.

From a Symmetrix perspective it is unlikely defragmentation will help much or at all with writes because unless you have a VERY heavy workload all of your writes are serviced by cache and reordered to the physical disks on the back-end.

However - for random reads which turn into cache misses you could see some performance gain if your OS blocks are defragmented resulting in multiple I/Os (or multiple random misses) going to the Symmetrix. Or even worse what should be a sequential read (i.e. reading the contents of a single large file) could turn into multiple "random" reads due to defragmentation at the filesystem level which could result in slower response times.

Running a defragmenation tool makes as much sense on a Symmetrix as it does on any other disks. Due to cached writes and larger buffer for reads compared to a regular drive you may not see as much impact as you would defragmenting your hard drive on your PC - but on a heavily fragmented system you will see some improvement.

Hope this helps.

1 Rookie

 • 

5.7K Posts

May 29th, 2008 07:00

Host based defragmenting = good :)
However: consider that when you have concattenated luns, the production data will be concentrated on the first members of your meta and thus the extra spindels in your meta aren't used. Striped metas are better for performance because of the fact that all spindels participate in all I/O's and if you defrag your data random I/O will be minimised because all data is sequential and all in 1 piece.

II hope you know what I meant to say since I am stumbling over my own words..... Tomorrow's weekend for me, you know ?

385 Posts

May 29th, 2008 12:00

You can't really "defragment" a Windows filesystem from the storage level because that would require knowledge of the filesystem structure.

That said - except in the odd case of the concatenated metas and the data being ideally fragmented to go across different drives - that hurts my head thinking about the odds on that - host based defragmentation could provide some performance boost.

The real answer is "it depends" because if the server is a database system the space is probably already preallocated and not subject to fragmentation (at the OS level at least) But for a file server or something where you end-up with lots of files coming and going a defragmenter could help.

3 Posts

May 30th, 2008 00:00

Many thanks, but all disk are striped metas.
Our Windows engineering believes they have a solution with disskeeper to improve the performance overall.
I thought all read/write operation goes over the RAM and I'm wondering how can deskeeper identify defragmentet file structure. ( pre chache function from symmetrix activated.)

3 Posts

May 30th, 2008 00:00

Many thank for your answer.
I'm absolute agree that diskeeper is not the right way to defragment a database like Oracle or SQL. But diskeeper germany wrote an article that diskeeper can help. http://diskeeperdeutschland.blogspot.com/2007_08_01_archive.html
No Events found!

Top