I have been fixing several bugs and finishing one RFE for FileBench. The specific improvements are:
Flowop deletefile has been enhanced to use an "fd" attribute, if supplied to locate the file to delete, rather than always picking an arbitrary one. To use this, you will need to close the file first, and then delete it before re-using the "fd" for another file. If "fd" is not specified, it deletes an arbitrary file just as before. Flowops no longer execute "slowly" while waiting for all threads to initialize. This was causing problems with workloads that iterate multiple times on one flowop. Instead, not flowops are executed until all threads have initialized. While some might view the former behavior as a sort of warmup period, the real solution is to add a real warmup period to FileBench, which is on a list of things to do. FileBench's semaphore flowops now save the semaphore descriptor when the semaphore is first created, rather than continually re-fetching it. This saves a rather expensive (from a cpu utilization point of view) system call from each use of the semaphore flowops. Modifed the remaining filebench_shm fields to have a prefix of shm_, so that they can be located easily with cscope. Fixed a problem with FileBench evaluating the "fileset" attribute too early, which caused it to fail if the fileset was not defined within the .f file before its first use. Now, like variables, you can define it later, perhaps in a separately loaded file, for more flexibility. The Web Review is located here: http://cr.opensolaris.org/~dreww/filebench_improv/ Comments welcome. Drew _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
