Re: RamFAST & GS/OS cache

[behrenss@hphalle6.informatik.tu-muenchen.de (Soenke Behrens)]

Todd P. Whitesel (toddpw@cco.caltech.edu) wrote:
: hal@budapest.math.macalstr.edu (Harold Byron Bouma) writes:
: >> I am going to bug Drew to support GS/OS cache :)
: >	Good, you should. Its much better caching and the GS will improve from 
: >it. Even the Turbo IDE card does it as well.
: Drew's position on this is that the method the RamFast currently uses is faster
: than it would be if he supported GS/OS caching. This is because the GS/OS cache
: has to (a) find the block by searching a list, and (b) manually move the block
: to the caller. So for data blocks the GS/OS cache is generally going to be
: slower than the RamFast unless you have a very fast Zip. But for directory
: blocks, the GS/OS cache makes a lot of sense _IF_ the FST's are smart enough
: to directly search for them in the cache rather than calling the driver and
: forcing a manual copy into the FST's own memory (which is also usually going to
: be slower than just blasting the block in from the RamFast).

Well, all I can give you is some experience: When copying a load of files, my
RF's performance drops. The Turbo IDE, which uses GS/OS caching (but only
for directory blocks and such) performs much better in that respect.

Soenke

--
===========================================================================
More than silver, more than gold, I need   !    behrenss@informatik.
electricity. I need it for my dreams.      !    tu-muenchen.de
(RACTER)                                   !  FIDO: 2:246/13 Soenke Behrens