Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!mnetor!uunet!seismo!ll-xn!cit-vax!oberon!castor.usc.edu!blarson From: blarson@castor.usc.edu (Bob Larson) Newsgroups: comp.sys.m6809 Subject: Re: 512K RAM upgrades Message-ID: <4132@oberon.USC.EDU> Date: Tue, 18-Aug-87 13:22:20 EDT Article-I.D.: oberon.4132 Posted: Tue Aug 18 13:22:20 1987 Date-Received: Fri, 21-Aug-87 01:04:29 EDT References: <5013@milano.UUCP> Sender: nobody@oberon.USC.EDU Reply-To: blarson@castor.usc.edu (Bob Larson) Organization: USC AIS, Los Angeles Lines: 42 Keywords: OS9 CACHE In article <1997@lsuc.UUCP> jimomura@lsuc.UUCP (Jim Omura) writes: > Well, I have to disagree with the trend here. I don't like full >cache for OS-9 systems. There are a *lot* of different approaches to >speed and system integrity. Full read/write cache adds speed but at >substantial risk of data loss and file corruption in the case of power >failure or other hardware flakey failure. Thanks guys, but no thanks. >Data integrity is too important for me to risk that. Read cache--which >was done by TLM on the Atari ST OS-9 port was, in my opinion, the >best compromise. Keep in mind that OS-9 allows you to *preload* >object files. This is much more efficient than cache. Secondly, >for datafiles, it's faster to throw intermediate files and reasonably >small files into a RAM disk. If a file won't fit in, say a 512K RAM >disk, a cache isn't likely to help that much either. A RAM disk has >most of the same problems for data integrity as a cache, but at least >you have an inherently better chance of retaining the "old copy" intact >even if you lose the copy in RAM disk. A partially updated disk copy >when the power goes down (or other "bad crash") is generally worthless. The chance of data lost between writing a properly designed disk cache and the flush to real disk should be extremly small. If I were designing one from scratch to go in a new OS, I would have the write to disk cache automaticly schedule the disk write with lower priority than a normal disk read. If the write had not completed in a reasonable about of time (10 sec or so) the priority should be raised. (Disk writes would happen when the disk would otherwise be idle unless the disk was being heavily used.) Explicit tempory files would be a nice idea to add to disk cache. They should only be written to the physical disk if there isn't enogh room in the cache. (Hopefully in such a way the space would be automaticly recovered on a reboot after a crash.) I consider ram disks a horrible cluge. They can only be used for explicitly temporary files or copies of perminant files. They do not automaticly change their contents based on what is being done. My ram disk on my qt+ is a source of problems: it's big enough to get in the way of memory intensive programs, but not big enough to hold the temporary files created by the c compiler on some programs. (I use a 128kbyte ram disk -- 512 kbytes ram on my system.) Bob Larson Arpa: Blarson@Ecla.Usc.Edu Uucp: {sdcrdcf,seismo!cit-vax}!oberon!castor!blarson "How well do we use our freedom to choose the illusions we create?" -- Timbuk3