Hi, ----- Original Message ----- | So about 18% improvement. That is worth having, but what about the read | side? What does the fragmentation look like afterwards as there is no | point in speeding up writes at the expense of reads.
Good question. Since the changes are confined to the GFS2 block allocator, which only affects writes, it should only degrade reads if fragmentation is made worse by the patch. But the patch actually improves fragmentation. Here's an example analysis output from 8 simultaneous iozone jobs: stock kernel: Fragmentation Summary /mnt/gfs2/iozonetest1: 17234 extents found /mnt/gfs2/iozonetest2: 17049 extents found /mnt/gfs2/iozonetest3: 17063 extents found /mnt/gfs2/iozonetest4: 17062 extents found /mnt/gfs2/iozonetest5: 17051 extents found /mnt/gfs2/iozonetest6: 16988 extents found /mnt/gfs2/iozonetest7: 17035 extents found /mnt/gfs2/iozonetest8: 17049 extents found patched kernel: Fragmentation Summary /mnt/gfs2/iozonetest1: 16622 extents found /mnt/gfs2/iozonetest2: 16613 extents found /mnt/gfs2/iozonetest3: 16611 extents found /mnt/gfs2/iozonetest4: 16616 extents found /mnt/gfs2/iozonetest5: 17315 extents found /mnt/gfs2/iozonetest6: 16611 extents found /mnt/gfs2/iozonetest7: 16612 extents found /mnt/gfs2/iozonetest8: 16656 extents found Regards, Bob Peterson Red Hat File Systems
