[Ocfs2-users] Much higher disk usage in OCFS2 then in XFS
Markus Meyer
markus.meyer at koeln.de
Thu Jul 17 01:06:37 PDT 2008
Sunil Mushran schrieb:
> That's 175 million files. I hope they are spread out across many
> directories.
We have them spread that there are not more than 10000 files per dir.
Hope that's enough.
> Our inodes are blocksized.
Ok, that explains that there is already so much space used after
formatting the disk.
> Maybe try 1K blocksize and 8K clustersize.
I played a bit with the settings and found out that a block size of 1k
and a cluster size of 4k is best for us. When I copied 1784 MB from an
XFS partition I had 1785 MB on the OCFS2 volume. Any other settings
didn't work since they ate up much more space(at least 80% more then the
original).
But now the write performance really sucks. I copy with an average of 2
MB/s. According to a bonnie++ test write performance(creating/deleting
files) is really bad. I added the bonnie stat below for anyone who is
interested.
> You would be an ideal candidate for the inlinedata feature we
> will release shortly. If possible, it stores the data in the
> inode itself.
Unfortunately "shortly" is not soon enough ;)
But thanks for your help nonetheless.
Cheers,
Markus Meyer
As you can see I used a lot of very small files. The box is an HP G4
with 4 3.4 GHz Xeon processor and is connected via an Emulex HBA to a HP
EVA 4400. So speed shouldn't be a problem.
# bonnie++ -n 100:100k:5k:120 -u 0 -d /stage/ -s 32104
[...]
thetin,32104M,39183,96,114672,90,71905,46,47594,86,177182,25,791.7,4,100:100:0,117,98,66781,98,7720,78,120,98,65725,97,357,95
More information about the Ocfs2-users
mailing list