Ian wrote:
>...or to implement a more efficient method of disk caching. Caching
>will necessarily slow down access when what you're looking for has not
>previously been loaded, but a good algorithm should keep the overhead
>to a minimum and not be affected greatly by the size of the cache. For
>frequently accessed blocks the benefits of efficient caching should
>outweigh the disadvantages - and reduce wear and tear on the disk.
Yes, of course. But if we wait for a comprehensive solution, we may not get
any solution at all. To this I very much prefer to remove SLAVEing, or
limits to it's memory usage. On a Q40 you can easily see the difference in
speed, if you use QDOS or Linux. Compared to this, you see what a crap
SMSQ/E is in this regard.
Modern harddisks have efficient cacheing methods of their own, and
remarkable cache sizes, so disk cacheing in RAM has become less important.
>An alternative for large memory machines is to use RAMdisk and have your
>most frequently used programs and data loaded in at boot time.
BTW simple loading is fast, because the "load" call isn't slaved.
The biggest problem are multimedia data or other data files that get random
accesses. With the current SMSQ/E you have no chance to write good
multimedia applications.
Maybe we must do something like this: Write a dummy device driver which
requests all memory available for slaveing, but doesn't use it. This way
the SMSQ/E IDE driver could not get memory for SLAVEing, because the dummy
driver had taken it before.
All the best
Peter
