Disk Defragger With EVA8000

[Robert Atkinson]

Given that (from what I've heard) an EVA8000 will move files around disks within the SAN, is it still relevant to run a Disk Defragger tool?

Rob.

[Robert Gezelter · Accepted solution]

Rob,

Disk defraggers generally work at the level of coallescing multiple extents of a file into a single, contiguous extent.

This is a level that the EVA-class devices are unaware. A disk controller (EVA or otherwise) can cache disk regions, and can deduce certain relationships among disk regions, but it CANNOT eliminate the XQP overhead in dealing with multiple extents (aka Window Turns).

How often you need to run a defragger is a totally different question.

- Bob Gezelter, http://www.rlgsc.com

[Hein van den Heuvel]

It's less relevant, but still somewhat.

+ a single file can have many extents and even many header to hold many extents. To worry about that is all overhead, no matter how little. THinking about those is all 'busy work' not working on the data itself.

+ as the system needs space for a new file, it is clearly easier to carve a single slice from a larger piece of pie versus gathering crumbs from all over.

+ with the EVA it makes even more sense to distribute files over disk based on use patterns, not by hard access rate or size:
1) You could have some disks where the files hardly move: large pre-allocate index files or database files come to mind. Other drives could be set aside for 'hopeless cases'. No need to defrag 1
2) lots of little files.. just let them fight amonst themself. Only occasionally defrag 2 for freespace
3) those large log files that stay open and grow and grow: hopeless. don't bother.

- EVA based disks tend to have larger cluster sizes, reducing fragmentation.

- EVA based disks will already involve mutliple spindles, so one you have to go to a different place on a different disk it is less relevant where on that different disk it is. And yet... if it is a large file, read frequently, spanning all drives coming back around to all drivers over an over then it is still 'nice' if most spindles can stay where they are and perhaps deploy read-ahead. (but other activity will readily disturb that).


- I am a firm believer in preventing.
- Too often I see 10x more resources wasted doing the defrag and the sub-optimal activity itself would take.
- Too often I see defraggers not achieving anything really.
+ The waste resource may happen at a time where there are resources to waste while the effects hopefully help at critical times where every overhead matters.


fwiw,
Hein.

[Robert Atkinson]

Thanks Guys.

I had an understanding of EVA's 'placement by useage' logic, but hadn't considered the potential of fragged headers, so will continue to run my defragger.

Rob.