Return to BSD News archive
Newsgroups: comp.unix.bsd.freebsd.misc Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.rmit.EDU.AU!news.unimelb.EDU.AU!munnari.OZ.AU!news.ecn.uoknor.edu!paladin.american.edu!zombie.ncsc.mil!news.mathworks.com!newsfeed.internetmci.com!in2.uu.net!news.thepoint.net!news1!not-for-mail From: root@dyson.iquest.net (John S. Dyson) Subject: Re: Linux vs. FreeBSD ... X-Nntp-Posting-Host: dyson.iquest.net Message-ID: <4n3sl5$ti@dyson.iquest.net> Sender: news@iquest.net (News Admin) Organization: John S. Dyson's Machine References: <3188C1E2.45AE@onramp.net> <4n0dhd$cff@agate.berkeley.edu> <31940EC3.61419F9C@lambert.org> <4n1hs2$shg@agate.berkeley.edu> Date: Sun, 12 May 1996 05:23:49 GMT Lines: 85 In article <4n1hs2$shg@agate.berkeley.edu>, Nick Kralevich <nickkral@america.CS.Berkeley.EDU> wrote: >In article <31940EC3.61419F9C@lambert.org>, >Terry Lambert <terry@lambert.org> wrote: > >>I have argued the danger of async writes at length in other >>threads. People's inability to accept the implications of >>"the completeness theorem" speaks only to their qualifications >>as scientists. > >(oh no, not the sync vs async metadata argument again. Ahhhhh!!!) > >Perhaps that's the difference between scientists and engineers. This >is just another example of the risk/rewards payoff. For volumes >mounted async, there is approximately 10X increase in speed vs async for >operations which do a lot of metadata update (such as untaring a file), >with a minimal (some would say non-existant) increase in risk. > The failing is in the *default* dangerous behavior of async writes used by Linux. I think that you are describing the difference between "(scientists and engineers) vs. cowboys." Whenever I use a filesystem that I am just playing with, I do use async metadata updates... That is appropriate for "playing" on FreeBSD or Linux. > >As you've pointed out, FreeBSD supports async writes. As you didn't >point out, Linux supports sync writes. Both options are available >on both systems. > Again, it is the default *broken* behaviour of Linux. Makes benchmarks look better, and you don't care if you crash during a benchmark, don't you? > >FreeBSD comes with sync writes on by default, where Linux has async >writes by default. I would argue that the Linux default is more >flexable than the FreeBSD default. Under the default Linux setup, >applications requiring sync access to the disk can always call >"fsync()", "fdatasync()", or "msync()", or open the file with the >"O_SYNC" option. > So, you are admitting that the Linux behaviour is more dangerous? Programs then have to do system calls to fix the broken behavior? The correct fix for the problem in many cases (but not always) is to use an J/LFS. You can still loose data though. > >It's up to the individual user to make the choice to take or reject >the risk vs reward of async updates. Lots of users take risks that >are much worse than this, such as running without a UPS or not making >backups. > Sure is, and I hope the user base makes the right decision. Remember, the correct decision is not necessarily always the fastest (especially when you have a system failure.) An equivalent argument is "why do backups -- they slow your system down?". > >Boy, you folks don't go out of your way to advertise this list, do >you? The list is not searchable in the FreeBSD search page >(http://www.freebsd.org/search.html), is not described in the >FreeBSD handbook under "mailing lists" >(http://www.freebsd.org/handbook/handbook251.html#459), nor is it >described anywhere on the FreeBSD WWW site. > Please refer to our mailing lists. Netnews is not our primary support mechanism (as we have stated over and over again -- being ignored by you.) Our primary support mechanism is not the handbook either. The MAILING LISTS ARE. Please redirect your questions to the appropriate forum -- you know how to do it. It is okay for new or casual users to use the net, but considering the questions that you are asking, and knowing what I know that you know, the mailing lists are obviously the best forum. > >>Does Linux support v10 of the POSIX threading, or only v4 of >>1001.3b? This is a serious question. > >I'll do some research and get back to you on this. > And tell me which release that it is in... Thanks John