sfio [was Re: mmap patch for Apache-1.0.5]

[John Heidemann <jo...@ISI.EDU>]

On Fri, 24 May 1996 14:11:28 EDT, Robert S. Thau wrote: 
>Ummmm... hate to keep harping on sfio, but it *does* do this sort of thing
>by default, on systems where mmap is available (among its other nice features).
>It would be interesting to see how performance with this patch compares to the
>unmodified distribution linked against the sfio stdio-compatibility headers
>and/or libs...

I think that sfio would give you:
	disk -> fs/vm-cache -> user buffer
	    -> mbufs -> network device
	(4 copies)
so you'd get something between the ISI_MMAP patch and the current
behavior.


I haven't been following Apache development for long, but it seems
like Apache-1.1's buff.c (a fairly small, Apache-specific stdio
replacement) fills a similar role as sfio (a presumably large, but
well-exercised and complete, stdio replacement done at Bell Labs).
I'm not sure you'd want both of them in Apache since their purposes
are so similar.  Which you want depends on how you trade-off size
vs. completeness vs. having-the-person-who-wrote-it-on-your-mailing-list.


(For those of you wondering what sfio is, the reference I have is:
@inproceedings{Korn91,
	author =	"David G. Korn and K.-Phong Vo",
	title =		"SFIO: Safe/Fast String/File IO",
	booktitle =	"{USENIX} Conference Proceedings",
	year =		1991,
	month =		"June",
	publisher =	"USENIX",
	pages =		"235--256",
}
).

   -John