It may not be "more" that you need. As Martin noted, the hashing algorithms generate a number between 1 and GSEMNUM based on the group address (and device/inode numbers), so that all records in the same group in the same file use the same semaphore set. Making this wider (incresasing GSEMNUM) does not affect that fact.
Type 19 and Type 1 files only have "group 1". These can be a bottleneck, so try not to use them for oft-updated data. Small modulus hashed files similarly can become a bottleneck if at critical points in the application, again as Martin noted. As a general rule, if you increase GSEMNUM, always ensure it is a prime number, and decrease GLTABSZ and RLTABSZ by the same factor. This means that the total demand for memory does not change by much. Tuning GLTABSZ and RLTABSZ involves monitoring LIST.READU at regular intervals for an extended period of time (at least two business cycles) then calculating the average and standard deviation of the number of locks held for each Lmod number (this is the number between 1 and GSEMNUM). The minimum depth (GLTABSZ or RLTABSZ) should then be set to μ+6σ (where μ is the mean and σ the standard deviation). Enrol on the UV910 (UniVerse Theory and Practice) class or the UniVerse Internals class (as it's still known in the UK) to learn more. ------- u2-users mailing list u2-users@listserver.u2ug.org To unsubscribe please visit http://listserver.u2ug.org/
