On Saturday, September 19, 2015 09:40:59 PM Tariq Saeed wrote: > and there is none, maybe there is one way down). This means looking at > lots and lots of chunks and each lookup involves a disk i/o. This is the > root cause of your problem. You should unmount and mount again > without specifying any value for local alloc or specifying 16 and you > will see a dramatic improvement in performance. > Regards, > -Tariq Saeed
Tariq - thank you! I brought the cluster down again today and remounted with localalloc=16 and I'm back to 700MB/sec writes. I'm wondering if this is likely to bite me again, though? It seems I may have screwed up by initially creating the file system with the default 4k cluster size. I could probably bring it down over Christmas and rebuild it with better settings if that's likely to prevent future problems. What would you suggest for mkfs options for an ocfs2 file system that's about 4TB in size and hosts only sparse virtual machine guest images (so very few, large sparse files, with unpredictable allocations and frequent hole punching)? _______________________________________________ Ocfs2-users mailing list Ocfs2-users@oss.oracle.com https://oss.oracle.com/mailman/listinfo/ocfs2-users
