Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.mel.connect.com.au!news.mira.net.au!vic.news.telstra.net!act.news.telstra.net!psgrain!usenet.eel.ufl.edu!tank.news.pipex.net!pipex!usenet1.news.uk.psi.net!uknet!EU.net!Norway.EU.net!nntp.uio.no!news.kth.se!tybalt.admin.kth.se!celsiustech.se!seunet!news2.swip.net!nike.volvo.se!cyklop.volvo.se!peter From: peter@cyklop.volvo.se (peter hakanson) Newsgroups: comp.unix.programmer,comp.os.linux.development.apps,comp.unix.solaris,comp.unix.bsd Subject: Re: How to flush write buffer ? Followup-To: comp.unix.programmer,comp.os.linux.development.apps,comp.unix.solaris,comp.unix.bsd Date: 13 Jun 1996 15:39:50 GMT Organization: Volvo Corp. Lines: 41 Distribution: inet Message-ID: <4ppco6$smm@nike.volvo.se> References: <4pg906$8su@sjx-ixn3.ix.netcom.com> <4phgrp$3hd@dawn.mmm.com> <4pigbq$qgb@dfw-ixnews3.ix.netcom.com> <31BEF990.51901C91@chromatic.com> NNTP-Posting-Host: cyklop.volvo.se X-Newsreader: TIN [version 1.2 PL2] Xref: euryale.cc.adfa.oz.au comp.unix.programmer:38471 comp.os.linux.development.apps:17605 comp.unix.solaris:71929 comp.unix.bsd:16802 This is the reason that databases should be run on raw-devices (and not on filesystems) Oracle used to have this possibility. There was also a performance gain (since oracle buffers internally, having a second buffer cache will waist cpu cycles ) peter håkanson Shankar Unni (shankar@chromatic.com) wrote: : Hugh McCurdy wrote: : > I must be using "logically" incorrectly or something. To me a : > "logical write" would be a write that would allow other processes to : > have access to the data written. : Absolutely. And even for buffered writes, almost all systems really *do* : treat the buffers as if they were a cache; i.e. if one program writes : the file (but the stuff is still in kernel buffers), and another program : requests a read from that file, the kernel will return the buffer : contents, not the disk contents. : The main reason for *synchronous* writes is to avoid problems with : system crashes and the like, where the kernel buffers are *not* written : back correctly. : For instance, many OSes do asynchronous writes for ordinary files by : default, but synchronous writes for *directory* updates, because those : data structures are sensitive, and an incorrect or partial directory : update can *really* screw up your file system.. : Typical candidates for synchronous writes (or at least, a synchronizing : operation) are database transactions (where you want to atomically : commit a transaction with some guarantee of completion), directory : updates, etc. : -- : Shankar Unni shankar@chromatic.com : Chromatic Research (408) 752-9488 -- -- Peter Hakanson VolvoData Dep 2580 phone +46 31 66 74 27