jlw@ariel.UUCP (11/21/83)
A week or so ago I posted a mild flame to net.unix_wizards lamenting the confusion that I saw on the net regarding the differences between a virtual memory system and a demand paging implementation of a virtual memory system. There were several responses on the net, some of then quite intemperate, implying that I was incorrect, and that, especially in academic circles, virtual memory was synonymous with paging. Therefore, I did some checking. I quote from a book actively used at Stanford at least up until two years ago. 'The term paging is used to describe a particular implementation of virtual memory and organization of main storage.' Shaw, Alan C., 'The Logical Design of Operating Systems,' pg. 117, Prentice-Hall, Inc. Englewood Cliffs, N.J., 1974. Madnick, Stuart E., Donovan, John J., 'Operating Systems,' pp. 139-186, McGraw-Hill, Inc. 1974. This reference, clearly, although not as succinctly as in the first example defines the three segment system of text, data, and bss (blank starting segment to answer another question), still used by BTL UNIX as an alternate form of virtual memory management. 'The general techniques of automatically moving required program and data blocks into physical M(ain)M(emory) for execution are called virtual memory techniques.' Hamacher, V. Carl, Vranesic, Zvonko G., Zaky, Safwat G., 'Computer Organization,' p. 252, McGraw-Hill Book Co., 1978. (all authors from Univ. of Toronto) I hope that this will answer any questions any of you may have had. Joseph L. Wood, III AT&T Information Systems Laboratories, Holmdel (201) 834-3759 ariel!jlw
mark@umcp-cs.UUCP (11/21/83)
Joseph L. Wood is still defending his use of virtual memory to describe
what BTL Unix does. Just a summary of the net opinion: don't believe
this guy on this issue.--
spoken: mark weiser
UUCP: {seismo,allegra,brl-bmd}!umcp-cs!mark
CSNet: mark@umcp-cs
ARPA: mark.umcp-cs@CSNet-Relaymark@cbosgd.UUCP (11/22/83)
I normally interpret "virtual memory" to mean that you can have more "memory" in a user process (transparently to the process) than there is physical memory on the machine. Games with overlays do not count since they are not completely transparent to the process. This does not imply paging, as there are other implementations (notably segmentation.) However, paging is probably the most popular implementation of virtual memory. That stuff you get out of BTL UNIX is perhaps best called "memory management", although there are implementations of System III or V from outside vendors that page.
jss@mh3bc1.UUCP (11/22/83)
Some computers map logical addresses directly to physical addresses some do not. Noticing the latter possibility was a major advance in the early days of computer science. While for many purposes the detatils of the mapping are significant the fundamental distinction is between no map (or the identity map if you prefer) and a nontrivial map. I have always used "virtual memory" to indicate a nontrivial map. A question for those who want to reserve "virtual memory" to describe particular maps. Do you have a name for the central concept? Jerry Schwarz BTL, Murray Hill
lee@fortune.UUCP (11/24/83)
I do not believe that these books are supporting your idea of virtual memory. 'The term paging is used to describe a particular implementation of virtual memory and organization of main storage.' Paging is certainly not the only virtual memory scheme. Segmentation is one of the others, but logical address spaces and overlaying schemes are NOT virtual memory. still used by BTL UNIX as an alternate form of virtual memory management. When they ( you know who ) started the UNIX development, real virtual memory schemes were not possible for the pdp-11. So they used the alternate form of virtual ( bad word choice ) memory management. 'The general techniques of automatically moving required program and data blocks into physical M(ain)M(emory) for execution are called virtual memory techniques.' BTL UNIX allocates all required Main Memory before program executions and ties up the memory blocks until program terminations. This is hardly automatic. IBM did not invent the name of Virtual Memory for nothing ( they didn't invent virtual memory though ). DEC would not build the VAX ( Virtual Address eXtension ) if they can sell the pdp-11 as virtual memory system. Ed Lee fortune!lee
tjt@kobold.UUCP (T.J.Teixeira) (11/25/83)
A small point:
Ed Lee (fortune!lee) says:
DEC would not build the VAX ( Virtual Address eXtension ) if they can
sell the pdp-11 as virtual memory system.
The biggest advantage of the VAX over the PDP-11 is the larger address
space you get with the VAX. Although you could build virtual memory on
the pdp-11, the 64K byte address space is too small anyway. Of course,
it is more important to implement virtual memory if you have a really
*big* address space.
--
Tom Teixeira, Massachusetts Computer Corporation. Westford MA
...!{ihnp4,harpo,decvax,ucbcad,tektronix}!masscomp!tjt (617) 692-6200root@basservax.SUN (12/03/83)
I must agree that System V doesn't have "virtual memory". But at least it will let you fill up your REAL memory (with something other than operating system). The versions 4.x of berkeley seem to want to keep 20% (or therabouts) of real memory free, which adds up to system death with overboard paging. Our set of active procs fit into real memory... Thus NO paging and 25% speed improvement (at least). Try this simple test. Enter vi, and then read vread(2) for about 30 seconds, and then type 'CR'. Now read the rest of vread(2) while waiting for your working set to come back...
ks@ecn-ee.UUCP (12/05/83)
#R:fortune:-183200:ecn-ee:13000004:000:787 ecn-ee!ks Dec 4 17:53:00 1983 In response to basserva!root... claiming that 4.x BSD swaps too much.. I have never noticed any real problems with swapping.. I have seen 60-80 users on a 4.1BSD system comfortably without noticeable swapping. Although, 110 users logged in did tend to kill it. It is true that system performance does drop significantly when swapping is necessary. I don't think I can ever recall a system around here swapping unless someone is using more virtual memory than the physical memory available. Perhaps your machine is scimpy on memory.. How much do have? I think most around here have 4 meg, with some heavily used machines having 8 or 16 meg.. The problems you mentioned have never happened to me. In my book, Berkeley is the only REAL UNIX for a VAX.. Kirk Smith Purdue EE