Details
-
Improvement
-
Resolution: Fixed
-
Minor
-
None
-
None
-
9223372036854775807
Description
For new writting, Lustre reserve credits in worst cases, eg for 1M write, we need reserve more than 500 credits.
This means we need much more journal space than data for each write, which means we need large journal to reach good performance.
One of potential way to improve this could be we allow transaction restart for write commit
case.
Attachments
Issue Links
- is related to
-
LU-14729 read-only remount due to transaction error in __ldiskfs_handle_dirty_metadata
- Resolved
-
LU-9985 osd_read_prep() ignores errors from osd_ldiskfs_map_inode_pages()
- Resolved
-
LU-14187 LU-14134 causes many racer timeouts
- Resolved
-
LU-14304 parallel-scale-nfsv3 test racer_on_nfs hangs
- Resolved
-
LU-14641 per extents bytes allocation stats
- Resolved