dillon@CORY.BERKELEY.EDU (Matt Dillon) (04/30/86)
Probably the best thing for AmigaDos right now would be a sector cache, implimented at the trackdisk level (or something like that)... You simply cannot make assumptions about the file system in a volitile system like this. The sector cache, of course, would not exist for the ram disk. The Amiga is notorious for not being nice to the poor user when it runs out of memory. I suggest another call be added that the system can make on devices, etc.... requesting them to get rid of any data they don't need (e.g. like a cache) because the system is running out of memory. For instance: The disk-driver can flush it's cache (when it gets one) The serial device can decrease the size of it's buffer Inuition can do lot's of neat things... like disallow window moves AND deallocate memory used for the purpose of window moves (I'm just guessing here) Most importantly, the system should return 'Fails' to user processes which request memory when there is less than, say, 10% memory left (that way, the system isn't likely to run out of memory at a critical time) -Matt
bruceb@amiga.UUCP (Bruce Barrett) (05/01/86)
In article <8604301613.AA07782@cory> dillon@CORY.BERKELEY.EDU (Matt Dillon) writes: > [...] > Most importantly, the system should return 'Fails' to user processes >which request memory when there is less than, say, 10% memory left (that way, >the system isn't likely to run out of memory at a critical time) > > -Matt :-) What a great idea! (-: ...except that "the system" doesn't know if it is the user or "the system" requesting the memory. Thank you but I don't think ignoring 891,289 bytes of RAM on a 8.5 meg machine is useful. The system, although not perfect does its best (so far) to exit from low memory conditions gracefully. For example windows stop moving and sizing. We are working to mak this even better and more robust. --Bruce Barrett
james@uw-june (James Synge) (05/04/86)
In article <1069@amiga.amiga.UUCP>, bruceb@amiga.UUCP (Bruce Barrett) writes: > In article <8604301613.AA07782@cory> dillon@CORY.BERKELEY.EDU (Matt Dillon) writes: > > [...] > > Most importantly, the system should return 'Fails' to user processes > >which request memory when there is less than, say, 10% memory left (that way, > >the system isn't likely to run out of memory at a critical time) > > > > -Matt > :-) What a great idea! (-: ...except that "the system" doesn't know if it is > the user or "the system" requesting the memory. Thank you but I don't > think ignoring 891,289 bytes of RAM on a 8.5 meg machine is useful. The > system, although not perfect does its best (so far) to exit from low > memory conditions gracefully. For example windows stop moving and sizing. > We are working to mak this even better and more robust. > --Bruce Barrett Agreed! We can't expect the system to successfully implement a policy we can not really figure out the details of ourselves: which programs are "system" and which are "user"? And when is the system asking for memory because of an action a user program asked for. Our programs must be VERY careful not to hog resources when they aren't really needed, and must check the returned results. To do otherwise is foolish and not very portable (like to a next generation machine). -- James M Synge