> > Yea, I remember some statements made about NAS's 14 way. > Something to > the effect of "any more than six"...However, wasn't this due to the > memory bus becoming overloaded? And isn't IBM's approach the use of > NUMA to overcome that problem? >
There are lots and lots of surprising scalability problems and they get harder as the individual engines get faster. The "other" platforms have an advantage in that for a great percentage of their work, they are effectively "shared nothing" systems. In other words, there is little or no interference between cpus (e.g. serializing access to storage) because the applications and middleware stack running on the engines tend not to "touch" the same stuff at the same time. That's the loose coupling model in a microeconomic scale. z/OS is resolutely tightly coupled. There are many many examples of resources that are hit repeatedly by all of the processors in the complex - hence the drop off in effectiveness as the number goes up. Removing those (real or accidental) scaling bottlenecks is one of the major priority items in z/OS design right now. CC ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
