virtual memory 4467
common segment" ... having started out as a 1mbyte "shared" segment in every address space. the hardware table look-aside buffers (TLBs) were STO (virtual address space) buttociative. the result was that there were...
On Fri, 12 May 2006 02:30:58 GMT in alt.folklore.computers, "kim
That's because most of the benefits are derived from having a large total address space to fit all your work. Of course, there are a number of different approaches to doing that. Read the ancient SUN paper on the benefits they got by changing their approach from managing real memory and paging space to managing a unified address space.
The VAX extension was from PDP-11 16 bit to VAX 32 bit. The page size dropped from 8KB to .5KB but the total real memory was initially the same 22 bits as the PDP-11, and the I-O was the same. They added a page modified bit but forgot about the page accessed bit. So OSes were constantly marking pages for page out and reclaiming them from the page out queue.
VMS implemented suboptimal local LRU, trying to guess a good working set size for the process, and then forcing the process to page if it needed to reference other pages, with various tweakable quotas per system and user. VMS also had an annoying tendency to have to drop the whole address space for various policy reasons, with much tweaking of quotas required to try to avoid this.
Unix ran on about 25% of VAXen to avoid those problems.
PDP-11 had virtual memory and swapping: 8 x 8KB pages in each 64KB address space (x 2 with split I-D space). Not such a big win.
Bumping the user's page limit, the system page space, an OS with a better algorithm, a machine with more memory would proabbly have been faster. And it's unlikely the machine was shut down: a ctrl-C (or two) should have toasted the process, and a ctrl-would have done the job.
precursor to as-400 was the s-38 ... which folklore has having been a bunch of future system people...
NT and newer derivatives still seem to use the old VMS drop the whole address space approach which leads to activity more similar to swapping than paging, and consequent delays, particularly with the bloated OS and apps, and vast quanbreasties of uncollected garbage.
kim kubik I have to agree with Nomad here, your explanation seems to add confusion. You dwell on thrashing but don't mention the...
Cheap enough for single user desktop use, but never cheap enough for any decent server load, particularly when desktop OSes are kludged for use as servers. Decent server OSes can be easily tweaked to provide good desktop performance: the inverse is much harder, as the design goals and consequences are so different.
Unix runs on a lot of servers to avoid those problems.
Brian Inglis ... It's (as usual) actually more complicated than that. While I can't remember exactly how VMS managed inter-process memory conflicts, NT and derivatives certainly attempt to do so with...
-- Thanks. Take care, Brian Inglis Calgary, Alberta, Canada
fake address use address above to reply
Alt Folklore Computers from Newsgroups