Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83 SMI; site creare.uucp
Path: utzoo!watmath!clyde!bonnie!akgua!whuxlm!spuxll!abnji!u1100a!pyuxww!gamma!epsilon!zeta!sabre!bellcore!decvax!dartvax!creare!gda
From: gda@creare.uucp (gda)
Newsgroups: net.unix-wizards
Subject: Speed in pipes
Message-ID: <357@creare.uucp>
Date: Wed, 12-Jun-85 22:20:17 EDT
Article-I.D.: creare.357
Posted: Wed Jun 12 22:20:17 1985
Date-Received: Sat, 15-Jun-85 10:21:33 EDT
Organization: Creare R & D, Great Hollow Rd,, Hanover, N. H. 03755
Lines: 29



Are pipes started by the shell faster than those set up by "popen", "fork
and exec", etc. ?

I've been building a three-generation program (parent-child-grandchild) in
which the three communicate in an ideal manner (no back-talk from the
children), and I tested the system previous to writing all the spawning code
by using explicit pipes in the command line :

		prog1 | prog2		(prog2 spawned prog3)

When I timed this, it ran reasonably fast, so I finished the code for
spawning of "prog2" by "prog1".  When I timed the final form it ran only
half as fast as when done with the explicit pipe.  I ran it under three
conditions in both "before" and "after" cases, one with a small amount of
input data, one with a lot, and one in-between.  There was a fairly linear
relation between data size and execution time.  I tried this with "popen"
and with "fork, exec, and dup2" and got roughly the same results.

This was done on a Masscomp, which is mostly System V.  I used "timex" to
measure the "user" time.  I don't have access to source code.

Any ideas?  Am I missing something?

			Gray Abbott
			Creare Inc.
			Hanover, NH
			{...dartvax!creare!gda}