At 10:58 AM -0800 11/18/97, Jonathan Shapiro wrote:
>Apparently I was unclear. No, I do *not* have in mind auto-rescinded pages.
>I was rather looking for a way to avoid the full context switch into the
>space bank by having a cacheing arrangement between the SB and the kernel
>that would permit the kernel to do quick response for some requests without
>altering the allocation semantics.
>
>>If I understand correctly, you are looking for a class of pages which are
>>automatically rescinded on restart. We never looked at those. I'm not
>>sure just how much less costly they would be.
We never looked at that. The 370 spacebank had the "feature" that there was a domain for each subbank and each subbank had to call it superior bank to allocate space. On the 370, one of the performance techniques was to minimize space banks. On the C implementations, we implemented a scheme where all the space banks were served by only two domains. One did all the allocations and retail deallocations, the other only handled space bank zaps.
Since we never had a chance to benchmark much with the C version, we didn't find out how much the performance improved. We obviously didn't take the next step of a kernel cache for allocatable space.
Bill Frantz | One party wants to control | Periwinkle -- Consulting (408)356-8506 | what you do in the bedroom,| 16345 Englewood Ave. frantz@netcom.com | the other in the boardroom.| Los Gatos, CA 95032, USA