I just migrated to a 120 from an 80. It was my first host-to-host Linode migration, and it was absolutely seamless. I was amazed how easy it was.

My question now is about the extra disk space that I now have. I want to use a chunk of it to store email, and I'd like to use maildir, which stores each email as a file. Obviously this leads to a large number of very small files.

How can I format the ext3 partition with a nice small cluster size? In the "Disk Images" setup, I can create a new ext3 partition, but that'll have the default 4KB cluster. I can create a raw partition and format it myself, but will I then lose the ability to resize via the Linode console? I don't think I'm willing to give that up.

Caker, any thoughts here? I tried using the raw partition and formatting it myself, just to see, and sure enough I couldn't resize it.

If I choose ext3 and reformat it myself with a smaller block size, will that work? Maybe that's a risky maneuver, though, if the resizer wasn't designed and tested with that in mind.

I guess I could always offload the mail someplace else and then back whenever I need to resize.

Please don't blame me if you don't wait for confirmation on this , but IIRC the resizer is just plain e2fsck + resize2fs. So I would recommend what you said last, adding an ext3 partition using LPM, then reformatting it with the custom block size. resize2fs should have no problem with it.

Makes sense to me. I'll probably still see if I can hear from the guy who set it all up, just to be sure.

He's right. That will work.

-Chris

I use reiserfs for maildirs on my office network. It's very quick and is probably the most efficient fs for small files (except for reiser4, too new for my tastes).

Only thing I don't like about it is when setting up a new filesystem (mkreiserfs) it doesn't check for bad blocks. You have to use the "badblocks" command to make a file containing the bad block mapping, then feed the file to the reiserfs tools. I've no idea how bad blocks relate to a Linode.

Update 4/29/08: John Dong (jdong) has enlightened me regarding the waning support of ReiserFS. The website of the developing company, Namesys, has now gone offline. OpenSUSE did have good reasons to stop using it as the default filesystem, and I wished I'd run across their reasoning earlier (link below). I'm reevaluating, and will probably start using Ext3 instead.

http://blog.linuxoss.com/2006/09/27/sus ... efault-fs/

I don't think the Linode will ever see bad blocks; each "partition" is just a file on the host IIRC.

You're right about Reiserfs and its nice handling of small files, especially since it doesn't have to round up to the cluster size. I had initially discounted it because I couldn't resize it from the LPM, but you've convinced me to reconsider it. I could always offload the data and put it back.

Found some benchmarks comparing mbox, mbx, and maildir on ext3 and reiserfs.

I will definitely be switching from mbox on ext3 to maildir on reiserfs!

http://www.decisionsoft.com/pdw/mailbench.html