It has been reported that there is a measurable performance improvement from increasing the maximum OST journal size from the current 1GB to 2GB or 4GB. It would be useful to measure the impact of this change (eg. IO500 run) and then determine if this larger limit should be made the default.

The mkfs.lustre code (via lustre/utils/libmount_utils_ldiskfs.c) is selecting the ldiskfs journal size based on the filesystem size, and controls the number and size of outstanding filesystem modification operations. However, it is undesirable to over-size the journal since this can consume up to an amount of RAM on the OSS equal to the sum of all MDT/OST journals on the server (including failover).

This limit is 4GB for MDT filesystems, but only 1GB for OST filesystems. It can be specified explicitly at OST format time with "mkfs.lustre ... --mkfsoptions='-J size=4000000' ...".