15-Sep-2008

Word for the week: Defrag

One of the novel activities here has been the manual use of BACKUP/RESTORE to defragment disk devices. Considering how it was done, it was rather dangerous. It was pretty easy to justify buying a license for DFO, based on the number of hours that the guys were coming in to defrag disks.

Over the weekend I scheduled some defrags for the production cluster, and to my suprise encountered some people that were how shall we say? uncomfortable about running the defragger there, even though it had been running in development for over a week without anyone noticing. I was asked for a test plan. This is rather difficult to accomplish for a product like this. Either is runs, or it doesn't.

Posted at September 15, 2008 3:14 PM
Tag Set:
Comments

Either it runs and your files are intact or they are scrambled egg.

Those people are probably remembering years ago when defrag products would do file moves in their own way and 'usually' worked. Since the movefile operation was added to the VMS file system I think they got a lot better.

However you have to ask - how much is the fragmentation impacting system performance in a user visible way and is it worth the downtime?

Posted by: Ian Miller at September 15, 2008 8:38 PM

Yep. I'd have to say that approximately 2400 window turns a second spread across four nodes in the cluster was a tad "excessive". On one of the disks I defragged over the weekend, the most fragmented file had in excess of 92000 extents.

Lesson for next week: teach the applications people about "pre-allocation" :)

Posted by: Jim Duff at September 15, 2008 8:55 PM

Comments are closed