Date: Thu, 5 Dec 1996 10:12:46 -0500 (EST) From: "John S. Dyson" <toor@dyson.iquest.net> To: se@freebsd.org (Stefan Esser) Cc: avalon@coombs.anu.edu.au, terry@lambert.org, lm@neteng.engr.sgi.com, thorpej@nas.nasa.gov, hackers@freebsd.org, torvalds@cs.helsinki.fi Subject: Re: The real issue... Message-ID: <199612051512.KAA04988@dyson.iquest.net> In-Reply-To: <Mutt.19961205145036.se@x14.mi.uni-koeln.de> from "Stefan Esser" at Dec 5, 96 02:50:36 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > [The Linux Byte Benchmark site (at www.silkroad.com) > offers pre-compiled Linux binaries, which give up > to twice as good results as building from sources > with no hand-tweaking on a current Linux-ELF system! > (They put statically linked versions of some commands > into the benchmark, which will be used instead of the > system commands in some tests. This is highly bogus, > it does not test system performance as is claimed, > but performance of those benchmark components. That > makes it even less useful than typical micro-benchmarks, > which typically let you know what meaningless metric > you measured :) ] > If one statically links processes that are forking or being execed, it would be best to specify it up front. The value in testing various programs for fork/exec time both dynamically linked and statically linked is that it gives you some hints as to the performance tradeoffs for each kind of binary. For example, almost NEVER dynamically link a command shell. Not only does it fool you by implying the program is smaller (it is smaller only on disk, most likely in memory it is larger, and sharing is NOT improved in non-degenerate cases), but fork/exec times are significantly higher. And of course, who wants an artificially slowed down command shell that fork/execs a couple of times more slowly than needed? IMO: It is legitimate to test in both the statically and dynamically linked cases, but the tester should disclose that information. John
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199612051512.KAA04988>