[Coco] Sugestions for new NitrOS-9 versions
lost at l-w.ca
Wed Apr 2 12:32:29 EDT 2014
On 14-04-02 10:17 AM, Aaron Wolfe wrote:
> RBF caching of sector 0 (read caching/write through) would give a decent
> performance improvement. I did some experiments a while back.. IIRC you
> can cache the stuff that gets constantly read using about 64 bytes so it
> doesn't even need a whole sector worth of ram. OTOH there may be a better
> solution to all those reads, caching individual disk parameters or whatever
> RBF is actually reading all the time. Maybe its just concerned that the
> disk has changed? Never got that far.
General block caching should be doable using free memory on the system
(if any). If the system needs the memory, cached data could simply be
dropped on the floor. This does have the issue of keeping track of what
is cached and from where, however.
For removable media, caching is dangerous, especially since we don't
have a "disk changed" notification from the hardware for floppy drives.
It gets even more dangerous if you cache writes.
For hard drives, however, there could be a substantial benefit. Perhaps
a device flagged as "fixed" or something could be cached.
The real problem is making sure that everything that accesses a block
device goes through the caching layer and making sure the cache stays
consistent in the face of writes. It's doable but not exactly easy.
I can see a significant benefit for folks with 2MB RAM.
More information about the Coco