home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.pc-clone.32bit
- Path: sparky!uunet!mdisea!kelsey
- From: kelsey@mdd.comm.mot.com (Joe Kelsey)
- Subject: Re: ISA Limitation?
- Message-ID: <1992Nov5.200108.6921@mdd.comm.mot.com>
- Sender: news@mdd.comm.mot.com
- Organization: Motorola, Mobile Data Division - Seattle, WA
- References: <1992Oct22.201908.5288@walter.bellcore.com> <1992Oct25.203602.2387@ksmith.uucp> <BwqMxn.26p@portal.hq.videocart.com> <913@felix.Sublink.Org>
- Distribution: na
- Date: Thu, 5 Nov 1992 20:01:08 GMT
- Lines: 36
-
- In <913@felix.Sublink.Org> eb@felix.Sublink.Org (Enrico Badella) writes:
- >In article <BwqMxn.26p@portal.hq.videocart.com> kdenning@portal.hq.videocart.com (Karl Denninger) writes:
- >>
- >>MAKE CERTAIN that your secondary cache scheme is set up to go beyond 16MB.
-
- >>If you have 64KB of cache on a '486 this is almost certain to be a problem.
- >>If you have 256KB cache it is probably not.
-
- >What are the reasons for such behavour?
-
- The problem really doesn't have anything *directly* to do with the
- size of the cache. A smaller cache generally means fewer cache hits
- and slower memory accesses. However, the important ``cacheable
- address range'' does have quite a bit to do with performance.
- Remember that a cache consists of two parts: the actual cached data
- (corresponding to the much-touted cache sizes referenced above) and
- what people usually call the TAG, or address part of the cache which
- stores the actual virtual address of the corresponding cache data lines.
-
- I just got through a very trying experience going from 16MB to 32MB on
- my 486-33 ISA machine. I ended up swapping out the original
- motherboard for a newer version. The old motherboard did not have a
- large enough TAG CACHE to handle beyond 16M address range. It did
- have 256kB of data cache, but the TAG CACHE couldn't handle more than
- 16M address range. My system booted and ran (Esix 4.0.4), but any
- programs which happened to get mapped into the upper 16M (common
- enough when you run a few large X Window programs (xv, emacs, etc.))
- had simply *horrible* performance, since the processor had to endure
- the full latency of every access to the 70ns rams!
-
- Now, I have a new motherboard with enough TAG RAM to cache the entire
- 64M address range of the board and the I can really *feel* the
- difference. XV simply screams along, when it used to crawl, the mouse
- responds much better, everything just runs better.
-
- /Joe
-