Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!unido!rwthinf!andrej!windy From: windy@andrej.informatik.rwth-aachen.de (Andrew John Stuart Miller) Newsgroups: comp.os.minix Subject: Re: Most requested features in MINIX Message-ID: Date: 5 Feb 91 15:19:25 GMT References: <43728@nigel.ee.udel.edu> Sender: news@rwthinf.UUCP Lines: 45 EOAHMAD%NTIVAX.BITNET@cunyvm.cuny.edu writes: >Am I alone? Are those people working on transputer and SPARC microprocessors >still insist on having virtual memory? Virtual memory is good provided we >have a standard chip to handle memory management, like serial and parallel >ports. Judging fromm Disk Controller designs, there will never be a standard >memory manager chipsets. It will always be constantly upgraded. I work with transputers (Helios) Sparcs (unix) Sun-3s (unix), PCs (Minxi) and STs (OS9-68000) It depends how you use virtual memory. For real time applications, anything that requires that data or programs be read from disk is unacceptable. OS9 takes this approach - programs are loaded and are not swapped out. to overcome the possible "Out of memory" problems, any program in main store exists as a module, and the text can be shared by many calls to that program. Some (complex and therefore not likely to be included in an offical minix release) software ensures that the store available is used as efficiently as possible. There are many problems which can only be solved without paging or swapping delays, and many other problems do not need to suffer these delays unnecessarily. For a transputer in a large network, the last thing I want is virtual memory to cause swapping or paging traffic through the network, or the cost of one disk per transputer, which would otherwise be needed. This is already a problem on all the sun-3 networks I have used, paging takes place via ethernet, rather than direct to a hard disk, and the delays are very noticable. Helios also tries to save memory in a similar way to OS9. There are many hardware and software solutions to memory management, just as for hard or floppy disk controlers etc. This is because different systems must meet different requirements. Andy Had one major requirement when mm and fs where implemented - simplicty.This leaves it up to the individual programmer to add his own things to a system when neded. Do not ask people to compilcate things unless completely needed..... Andrew Miller -- -------------------------------------------------------------------------------- email: windy@strange.informatik.rwth-aachen.de snail: Ruetscherstr 165 D-5100 Aachen voice: 0049 (0)241 894-355