Hey all.... Any ideas on how to do a defrag on a Red Hat 7.5 system?

We have some issues with a couple of frag'd drives and it's killing performance.

 

I thought you dont have to defag linux file systems?

 

Ya' know what's fun... all the info on the 'net that says that is wrong. Seriously.


We've just spent the last 36 hours recreating a field problem here in our lab.

I need to be somewhat vague, but... We have a customer that's insisted on not deleting any of the content files that they've gotten at their (700ish) stores over the last couple of years. That amounts to about 12,000 files, and our control files are over 16mb each. This is on a 36gig partition, and the net result is that the drive is about 93% full.

We can _clearly_ show that there's a threshold where ext3's housecleaning (this is why you "don't have to defrag a linux file system" ) starts to fail because it simply doesn't have enough space on the drive to defrag files. We have frag tools that clearly show the amount of fragmentation on the drive and within specific files that are important to us.

The net result in our application is a situation that doesn't make the customer happy. Rightly so, but the customer is pushing the system past the line that we've recommended.

The net net result is that myself, my cohort in coding crime and our tester have been jumping through hoops to "fix" this remotely.


We're confident that we know what's going on, but we'd love to find a defrag tool to be able nail this down and have solid proof to our management.

 

you shouldn't need to defrag on linux/unix machines, but it can become a problem. You may want to make a new partition, copy all files to that drive, format your storage partition and then copy the files back. Good luck

 

wow... old thread.

To update, we wrote some tools to force _drive_ fragmentation and dutifully watched ext3 fail in keeping things neat and tidy. It tried, but simply didn't have enough free disk space to make it happen - all the while though it killed performance on the boxes.

We also wrote a tool to force _file_ fragmentation for a couple of our key system files (VERY large xml nightmares) and stood by while our system choked up but good.

We've since reworked this old file manager code and things are a lot better. It's still interesting to note that while "we shouldn't have had to..." we most certainly did have to.

The file copy/repartition/file copy method would have worked here in the lab, but we needed something that we could deploy to the field. We have 20k+ machines in the field that are standlone. About 700 of those are at 90% capacity and we're (now) about a month away from them being in a critical state such that we have to really understand what happened to the small test sample so that we can remotely deploy a global fix.


weeeeeeeeeeeeeeee!