[Linux-cluster] Question about GFS2 journal size
rossnick-lists at cybercat.ca
Tue Dec 13 21:18:19 UTC 2011
> For GFS2 journals, you don't need to create extras like GFS1.
> You only need one journal _per_mounting_node. So your cluster
> can be 32 nodes, but if only two of those nodes are going to
> mount the file system, you only need 2 journals. So the question
> is how many of your 8 nodes are going to mount?
> If all 8 need to mount it simultaneously, then create 8 journals.
> With GFS2, you can always easily add more journals with gfs2_jadd
> (unlike GFS1).
Thanks, I'll then only make 8 journals...
> Now as to journal size: The biggest factor to consider here is
> performance. The reason you want big journals is so that the
> processes writing metadata to the file system don't have to pause
> and wait for journal flushing very often, which forces the journal
> to be synced to the media, which is slow. IIRC, the minimum journal
> size is 32 (MB), so 16 isn't an option. You probably wouldn't want
> 16MB anyway, because it would impact performance, as explained above.
> So it depends on how you're planning to use the file system.
For the particular instance I am building, it's a "test service" that
doesn't require almost no writing at all to the FS. And as for the
minimum size, from the man page, it's 8 megs. So for this one, i'll go
with 16 megs, that'll be enough...
More information about the Linux-cluster