Subject: Re: Packaging Linux From: rmk@rmkhome.UUCP (Rick Kelly) Date: Mon, 23 Nov 1992 07:15:30 GMT
In article <1992Nov20.190818.27245@samba.oit.unc.edu> jem@sunSITE.unc.edu (Jonathan Magid) writes:
>In article <1992Nov20.160444.16116@scifi.uucp> njs@scifi.uucp (Nicholas J. Simicich) writes:
>>In article <17612@mindlink.bc.ca> Neklan_Brozensky@mindlink.bc.ca (Neklan Brozensky) writes:
>>
>>I personally think that another essential for mass-usage is going to
>>be a filesystem that doesn't require sync() or fsck. This isn't an
>>impossible task, and yes, there is a performance hit, but I don't
>>think you are going to sell a system to the masses that they can't
>>turn off whenever they want to.
>>
>This isn't really a function of the filesystem so much as U*IX-like os's
>standard use of buffer caching to speed disk access. I think that the
>performance hit for disk intensive jobs is higher than you think...
>(try dyking out the buffer cache code in the kernel, if you want to try,
>or easier put sync in the crontab to go off every second, come to think
>of it, thats a way to make a system for people who insist on no buffering)
SVR4.2 with the Veritas file system does a sync() on last write. Univel
claims that a secretary will be able to type a letter in Word Perfect, save
the file, and then hit the power switch. I believe that Veritas is a
journaling file system, and they obviously have sync() written to return
on successful execution rather than successfully being scheduled.
Maybe someone will reverse engineer the Veritas FS. :-)
--Rick Kelly rmk@rmkhome.UUCP unixland!rmkhome!rmk rmk@frog.UUCP