Duration and impact of online resize of a large(ish) file system

Latest response



I recently did a resize of a relatively large file system (ext3 on RHEL5), as described in this ServerFault thread: http://serverfault.com/questions/410325/online-resize-of-large-ext3-file-system-is-performance-affected


When trying to plan for duration and performance impact of increasing a 4TB filesystem with 1TB, I first did a read-only fsck, which ran for 5 hours. Therefore, due to advice that resize2fs will take 1.7-2x the time of an fsck, I budgeted 8-10 hours and running the resize in a "quiet" window. However, when I finally did the resize, it completed in 20 minutes.


Does anyone have insights on why it completed so much quicker than the read-only fsck, and how to plan for similar actions in the future?

Is there or not reason to worry about an performance impact while doing online resizing?


best regards,




Doing an resize2fs to *reduce* the size of a fileystem typically takes the time-factor you cite in your post. That time factor is resultant of  recalculating the entire filesystem and redistribution of fragments and data-blocks to fit within remaining contiguous space.


When upsizing, your resize will depend mostly on how much you're resizing by. The resize is mostly just creating the necessary filesystem structures (and pointers) to reference the newly-available storage blocks. That's a lot faster than the operations necessary for a reduce.


On a related note, if you're frequently going to upsize filesystems, you might want to use an extent-based filesystem. Upwards resizing is nearly instantaneous.

OK, thanks. We're usually not doing this kind of resizing - for most other Oracle DBs we give disk directly to Oracle ASM. In this case we're using ext3 as ext4 was not officially supported by Oracle at installation time.



Given the recent issues with ext4 data-integrity that's made the tech press (Google "ext4 corruption bug" - you'll get a number of hits from late October of this year), I don't see Oracle doing that any time soon. Add in Oracle's agenda for pushing ASM and eliminating any use of third-party filesystems, volume managers or clusterware, and ext4 support seems likely only if a very large customer demands it of them.