--- "Jeff Gribbin, EDS" <[EMAIL PROTECTED]> wrote: > With a small amount of trepidation (but inviting > stomping from anybody who > feels that I'm off-base here) can I remind folk > that, on IBM mainframe > hardware, MIPS aren't the whole story. There's > channels too - and in an > I/O-related situation their power needs to be ADDED > to the CPU power to > come up with a realistic, "comparative MIPS" figure. > > It's a very long time since I saw anything that > indicated how much MIPpage > is offloaded into the channels by a typical, > "mainframe" workload but > please remember that, unless you understand how > channels are implemented > when comparing two different solutions, you can > quickly mislead yourself > regarding the genuine value of the, "MIPS" > comparison. > > (I have a similar problem regarding, "channel > bandwidth" - each individual > channel on a mainframe might be, "slow" but > potentially I can have several > hundred running in parallel - in the right > circumstances doesn't this give > me greater capacity to work with than a single but > much faster I/O portal? > Do I want a firehose or do I want the Mississippi? > As a man to whom I > would happily defer when it comes to performance > issues has occasionally > been heard to comment, I think, "It depends ...") > > Regards > Jeff Gribbin (Speaking only for himself.) >
Jeff, Hercules runs channel emulation and CPU emulation in separate threads, so in a multi CPU box with say "n" CPUS, if you define "m" Mainframe CPU, "n-m" are generally (pedants note generally) free for channel emulation. However whilst I have never tried to do a real benchmark, I am firmly convinced that I/O is not an issue on a modern PC. To expand a little, I have tried a few simple things to drive the I/O system up and bottleneck the I/O in Hercules.. Sadly, every time, I have failed. I do keep trying, but I have never been able to justify adding RAID, SATA, or even SCSI (other than for tape) to the box I use for Hercules. When I look in PERFMON the i/o queue length and the i/o service times remain short. As I only emulate one CPU and have (kind of two) on the Hyperthreaded box, I see the second CPUs utilization remains low. I have therefore concluded that emulating S/370 channels does not tax the system. Again it might be different for the XA I/O system , but I don't think so. (In fact I think it may be simpler) Dave. Also speaking for himself. ____________________________________________________________________________________ Do you Yahoo!? Everyone is raving about the all-new Yahoo! Mail beta. http://new.mail.yahoo.com
