DDR also uses synchronous IO, so no overlapping possible. Threading will not help to make DIAG I/O's multithreaded: CP suspends the virtual machine during the DIAG I/O. To be precise: there is a newer DIAG (250) to make asynchroneous I/O, but that is not used by CMS. If you wonder why: exploiting asynchroneus I/O is much more complicated, and CMS' nature was single user. This newer DIAG is there for Linux.
2010/7/26 Leland Lucius <lluc...@homerow.net> > Rob van der Heij wrote: > >> On Mon, Jul 26, 2010 at 7:57 AM, Leland Lucius <lluc...@homerow.net> >> wrote: >> >>> Is there anyway to do it? I've tried using MTREXX, but the threads run >>> after each other. I've tried multiple stages, but they also run after >>> each >>> other. >>> >> >> IIRC the trackread and trackwrite stages use diagnose I/O under the >> covers, so would block there. Depending on what happens in the rest of >> the pipe, it may make sense to build a more complicated topology (like >> with long haul IP connection with varying latency). But probably not >> with CMS MT as you suggest. You probably should run multiple virtual >> machines and break up the work like that (though if the disk is the >> bottleneck, you're not likely to achieve much speed-up). >> >> Yepper, I'm to the point of just doing the single threaded, but, even if > the trackwrite used a DIAG, then I'd expect threading would get a chance > between DIAG calls. I've tried adding a ThreadYield() in a REXX stage, but > that just causes all sorts of problems. > > Just out of curiosity, I think I'll try DDR before throwing in the > towel...though I'm not gonna try too hard. :-) > > Thanks much, > > Leland > -- Kris Buelens, IBM Belgium, VM customer support
