Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have a system that has a high throughput of small files on disk, i.e. a huge number of small files are created, written and deleted within seconds.

Are there any reasonable ext2/ext3/ext4 mount options to improve the performance? I guess, meta-data journalling results in a huge performance drop here.

share|improve this question
up vote 3 down vote accepted

Yeah, metadata ops are going to absolutely kill you. The most important mount option I can imagine helping would be noatime, which turns off atime (or "last access time") updates on all files. That'll stop one metadata write for every file access, which might halve your I/O rate (if you read/write each file once) to hundreds of times (if you write once, read many). noatime also implies nodiratime, which turns off atime updates just on directories. If that's a bit too brutal (you need atime sometimes), then consider relatime (mount(8) explains that one better than I can).

On a hardware level, seriously consider more RAM and a non-volatile cache hardware RAID controller. More RAM helps the kernel cache more data, which reduces (or can even eliminate) read I/O, and the NVRAM cache RAID controller means that your data is safe and secure after it's been written to flash (which is fast), rather than all the way down to the spinning disks (which is slooooow). You can also go SSDs, but they're still significantly slower than NVRAM.

share|improve this answer
Nice answer. What about turning off journalling (via tune2fs -O ^has_journal) or using ext2. Would that improve performance? – Michael Aug 7 '12 at 9:28
I haven't benchmarked those, but I'd be wary of them because they reduce durability. If you can trade-off durability for performance, there are a whole range of interesting things you can do... – womble Aug 7 '12 at 9:30

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.