On Tue, 5 Feb 2008 07:09:56 +0000, Ted MacNEIL wrote: > >When I was responsible for a beta test of MDF, when it first came out, >I was told that the cache line was split evenly across domains. And, it >was partitioned and dedicated to each domain.
I don't know what you mean when you say the cache line was split across domains. I forget whether a line was 32 bytes, but it always cantained the data from a line of storage. The cache was split into an instruction cache and a data cache, though. There was significant logic needed to deal with that. Further partitioning it by domain would have been an engineering nightmare to make it work well. To chop up the cache into 16 segregated caches would have left each domain with very little cache. And there would have needed to be one for macrocode as well. For some domains there would have been insufficient cache while others would have more than they needed. How would it be divided? Proportionately to main store size? Equally? What happens when a domain is deactivated? When a new one is activated? Indeed, cache was one of the reasons the domains were dispatched by time slice. A domain was allowed to run for a while before another one was dispatched. >This was from very experienced Amdahl reps. > >.. But, if you know better ..? Well, that was a long time ago and I don't have any documentation any more, but I was a 580 specialist. I made it my business to understand how the architecture worked. Not everyone in Amdahl had access to the ALTA POO. I had it, I read it and I understood it. -- Tom Marchant ---------------------------------------------------------------------- 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
