Please don't reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10816
(In reply to comment #19) Some comments from the Cray testing group about the test plan. Given that the purpose of the allocator is to reduce fragmentation (and give better and more consistent performance?) on aging filesystems the test plan seems to be lacking in any fragmentation and performance measurements, particularly in regard to a filesystem with significant longer term use. There should be some mention of the impact of this change on the performance of metadata operations and how it will be measured. The plan doesn't mention running CFS's regular regression tests. It would also be nice to have the list of other workloads and tests CFS plans to run. There should be more description of hardware being used. It should include running on a raid. It should include running on at least 2 TB LUNs. There is nothing about the interaction with lustre striping: stripe sizes, stripe counts, etc. There should be testing of a large file system - multiple osts, using multiple multi-processor jobs, doing I/O to multiple, differently striped directories. There doesn't seem to be any checking of data integrity as part of the tests. Add segmented and random I/O to the testing. Are there tunable allocation parameters or policy choices? If so, they should be tested. Estimates of effort and schedule? Who is responsible for the testing? How and to whom will results be presented? What is not being tested and what are the associated risks? This is important to us to know as it gives us some idea of what areas may be weak and we can decide how to mitigate the risks. _______________________________________________ Lustre-devel mailing list [email protected] https://mail.clusterfs.com/mailman/listinfo/lustre-devel
