I have completed all the updates after the last review comments and completed all my testing. I have done manual tests and Filebench tests on SPARC and x86. The Filebench tests were also repeated to kmem_flags set to 0x1f and checked for correctness and leaks. The fsfuzzer test was run for 10days without issues. The bug is updated.
The updated webrevs are available at: http://www.genunix.org/distributions/belenix_site//binfiles/hsfswebrev/index.htm Updated binaries available from: http://www.genunix.org/distributions/belenix_site//binfiles/hsfs.amd64 http://www.genunix.org/distributions/belenix_site//binfiles/hsfs.sparcv9 This update adds the following changes: - Incorporate all the review comments. - Add kstats and SDT probes for debugging purposes. You can copy the updated module, mount a DVD and see: kstat -m hsfs_fs dtrace -l -P sdt | grep hsfs - SPARC testing was done on a T2000 and it threw up a surprise perf issue for the single-stream sequential read and random read tests. It was found using standard kmem_alloc for thousands of small allocations of size 96 bytes was inefficient. So a kmem_cache was used and it made a dramatic difference. - Removed one unnecessary page_lookup in the readahead path. As it stands now there is an across the board improvement on all Filebench tests including the random read tests on both x86 and SPARC. Regards, Moinak.
