On 02/06/2014 05:42 PM, Shmuel Metz (Seymour J.) wrote: > In <bc6b7b77ef4d5e4bb95b0559a0d6d5e10531ce9...@mx3.state.nv.us>, on > 02/06/2014 > at 01:17 PM, "David G. Schlecht" <dschle...@admin.nv.gov> said: > >> Or am I way off base? > > At least partially. On a simulated 3390 a cylinder boundary may not > tell you much about seek time, but it can still have an effect on > whether a channel program breaks and has to be restarted. OTOH, with > code that builds an ECKD channel program that has a lot less impact on > performance than it used to. > > ...and if the physical storage media still has an access-time function which tends to be larger when separation on the physical media is larger, it is not unreasonable to expect that any emulated 3390 strategy which utilizes a fixed mapping would likely use some sequential allocation strategy which would at least show a statistical bias for faster access to blocks that were closer together on the emulated device, as that would improve the odds (but not guarantee) they are closer together on the physical media as well. Such access-time bias would no doubt vary in a manner that couldn't be usefully predicted from emulated DASD architecture, but it seems reasonable that it should exist. In a practical sense though, one could argue that any data that is being accessed frequently enough to be a performance concern would be interacting with DASD cache storage in ways that could completely mask the difference in physical access times; so the controlled placement of VTOC, VTOCIX, and VVDS these days is more a matter of aesthetics and emulated-media fragmentation avoidance than performance.
-- Joel C. Ewing, Bentonville, AR jcew...@acm.org ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
