Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.mira.net.au!news.vbc.net!samba.rahul.net!rahul.net!a2i!olivea!spool.mu.edu!howland.reston.ans.net!Germany.EU.net!Dortmund.Germany.EU.net!interface-business.de!usenet From: j@ida.interface-business.de (J Wunsch) Newsgroups: comp.os.linux.misc,comp.unix.bsd.freebsd.misc Subject: Re: TCP latency Date: 25 Jul 1996 10:35:07 GMT Organization: interface business GmbH, Dresden Lines: 42 Message-ID: <4t7ikr$g42@innocence.interface-business.de> References: <4paedl$4bm@engnews2.eng.sun.com> <4s8cuq$ljd@bertrand.ccs.carleton.ca> <31E7C0DD.41C67EA6@dyson.iquest.net> <4s8tcn$jsh@fido.asd.sgi.com> <31E80ACA.167EB0E7@dyson.iquest.net> <4sadde$qsv@linux.cs.helsinki.fi> <31E9E3A7.41C67EA6@dyson.iquest.net> <4sefde$f0l@fido.asd.sgi.com> <4socfr$3ot@dworkin.wustl.edu> <4t63as$f2q@fido.asd.sgi.com> Reply-To: joerg_wunsch@interface-business.de (Joerg Wunsch) NNTP-Posting-Host: ida.interface-business.de X-Newsreader: knews 0.9.6 X-Phone: +49-351-31809-14 X-Fax: +49-351-3361187 X-PGP-Fingerprint: DC 47 E6 E4 FF A6 E9 8F 93 21 E0 7D F9 12 D6 4E Xref: euryale.cc.adfa.oz.au comp.os.linux.misc:118365 comp.unix.bsd.freebsd.misc:24370 (Removed the NetBSD group per their request, they don't seem interested in benchmarking. ;) Moved the linux group to .misc, hope that's the right one.) lm@neteng.engr.sgi.com (Larry McVoy) wrote: > kill(), umask() These are the best "null system call" benchmark > choices I've seen. I'd vary the mask in the umask > one so that it was actually changing state. The Well, the kill() proposal is ``not quite null'' either, since it has to walk the process table. If the process table is linear, finding process 1 will certainly be quick, but they are often (usually?) hashed, so the result is probably as accurate as for the /dev/null write: you've got a good measure to see whether some optimization on your own system will make things better or worse, but the numbers hardly compare to other systems (since they have no real meaning, just like the ``write 1 byte to /dev/null'' is far off from real-world usage patterns). umask(umask(0777)) does indeed always change state, but the effect of setting the umask is IMHO so that any attempt to `optimize' the `doesn't change anything' case away will probably be a pessimization, i.e. even setting it to the previous value is cheaper than launching a comparision (and an expensive branch) first. > I'm also willing (and interested) to find a different syscall that just > measures trap overhead, but I haven't seen one yet that I really like. > The getppid() may be the best out there, though, it's hard to cache > that. Thoughts? Other people also suggested getppid(). I'd say either this or the umask() were my favorites. Btw., if you decide to have a ``null IO syscall'' test, why not also comparing 1024 bytes to /dev/null as well? This would allow an estimation of scaling problems. -- J"org Wunsch Unix support engineer joerg_wunsch@interface-business.de http://www.interface-business.de/~j