home *** CD-ROM | disk | FTP | other *** search
- Path: felix.teclink.net!usenet
- From: rad@techlink.net (rad)
- Newsgroups: comp.sys.amiga.misc
- Subject: Re: OS features
- Date: 10 Jan 1996 02:11:02 GMT
- Organization: TECLink Internet Services: info@TECLink.Net
- Message-ID: <2530.6581T1360T1779@techlink.net>
- References: <4cf7vs$gfk@ArtWorks.apana.org.au>
- NNTP-Posting-Host: tc1_42.teclink.net
- X-Newsreader: THOR 2.22 (Amiga;TCP/IP) *UNREGISTERED*
-
- >Christopher D. Judd (cdj@gmk02.chem.rpi.edu) wrote:
- >[...]
- >-> Virtual memory is fairly trivial. It's already been done (VMM, etc.)
-
- >I wouldn't say trivial (for an existing operating system) but certainly
- >doable. VMM & GigaMem are proof.
-
- >-> so users who have slower processors or just don't want the performance
- >-> hit don't have to use it and those of us to whom stability is more
-
- >This is a common misconception. There is no significant performance hit
- >with virtual memory, the address translations are done in hardware
- >effectively as part of the RAM. If they couldn't be done within a RAM cycle
- >then somebody would find a faster way of doing it (read wider RAM).
- >So there is no performance hit.
-
- You are missleading people here. Virtual memory causes a small performance
- hit over running in ram for three reasons:
-
- 1) Address translation can only be done within regular RAM cycles on a ATC
- (address translation cache) hit. Just like in the operation of caches,
- ATCs can only hold a few address translations internally (64 in the case
- of the 68040). On a ATC miss a page table search in main memory (upto
- three reads) must be done. Sometimes a modified page descriptor must be
- stored back to memory to make room for a new entry in the ATC, this could
- involve up to three additional read or write cycles.
-
- 2) The ATC is a write through cache and since the ATC contains a page-
- modified bit which needs to be updated on the first write of on a page,
- their will be additional (up to three) additional read or write cycles
- even though the descriptor may already be in the ATC.
-
- 3) Since a some ram is required to hold page tables you will on occasion need
- to you swap space under virtual memory you would not need under a non-
- virtual memory system.
-
- >You do however get the appearance of
- >more RAM, the extra RAM is slower than adding more RAM chips, but to
- >call that a performance hit is misleading. Rather it is extra functionality.
-
- You are correct that VMM adds an important functionality. (Some programs
- that could not be run without out it, will run with it.)
-
- ------------------------------------------------------------------------------
- - Richard Deken EMail: (personal) rad@teclink.net -
- - VLSI design engineer (AuE business) rad@aue.com -
- - Advanced Microelectronics PGP public key available -
- ------------------------------------------------------------------------------
-
-