This is the mail archive of the cygwin@sourceware.cygnus.com mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: Time and motion studies of gcc and egcs and LCC


> Discarding pages in strict FIFO order seems to me like a recipe for
> thrashing.

     I don't know about Win95, but VMS and (its closely related 
descendant) NT make no attempt to maintain a working set in the 
conventional (Denning*) meaning of the phrase:

	pages are trimmed from processes when the system gets
	short of memory, but not on any kind of LRU basis

	these pages are sent to a kind of limbo list* with
	their contents intact

	if a page faults and is on the limbo list, it's
	given back to the process (this is called a soft
	fault and is MUCH cheaper than having to read the
	page from disk)

	pages that stay on the limbo list too long end up
	getting reused for new pages

Obviously there's a tradeoff between the overhead of maintaining some 
kind of LRU ordering and the overhead of taking lots of soft faults.
Does anybody know of any studies on this tradeoff?

     If the perfect is the enemy of the good and the good is the enemy 
of the good enough, the soft fault approach is probably good enough.  
However, my experience with software that strongly exploits VM (e.g., 
Lisp environments) is that such software performs much better in a 
conventional working set environment (say, Twenex) than it does in a 
soft fault environment (say, VMS).

*see http://paris.lcs.mit.edu/~bvelez/std-colls/cacm/cacm-1751.html,
 for one

*NT calls this the standby list - go to the Performance Monitor, 
 select Object: Memory Counter: Page Faults/sec and hit Explain.
 Also, AltaVisting for "standby list" and "soft fault*" turned up:
 http://www.tek.com/VND/Support/Release_Notes/PERFMON.html and
 http://www.developer.com/reference/library/0672309777/mbaxc.htm
-------p--a--s--s--i--o--n-----n--e--e--d--s-----a-----f--a--c--e-------
"oncology recapitulates philately" --Mark Maxson    Robert M. Praetorius
"balance, not symmetry" --Mark Stanley    work: RPraetorius@AspenRes.Com
(attribution by Stigler)     fun & recreation: rmp@PopJ1.MA.UltraNet.Com
-
For help on using this list (especially unsubscribing), send a message to
"gnu-win32-request@cygnus.com" with one line of text: "help".


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]