Path: utzoo!utgpu!water!watmath!clyde!burl!codas!cpsc6a!rtech!llama!daveb
From: daveb@llama.rtech.UUCP (Dave Brower)
Newsgroups: comp.unix.wizards
Subject: Re: Unix vs. OS/2 (was RE: Pournelle on Unix)
Message-ID: <1566@rtech.UUCP>
Date: 12 Jan 88 03:42:35 GMT
References: <11156@brl-adm.ARPA> <10091@mimsy.UUCP>
Sender: news@rtech.UUCP
Reply-To: daveb@rtech.UUCP (Dave Brower)
Organization: Relational Technology, Inc. Alameda, CA
Lines: 22

In article <10091@mimsy.UUCP> chris@mimsy.UUCP (Chris Torek) writes:
>In article <11156@brl-adm.ARPA> GUTHERY%ASC%sdr.slb.com@RELAY.CS.NET
>>	- file write-though
>
>As for these, I am not sure what is meant...

A critical issue to database types:  Making sure the contents of a
write actually hit the disk.  Fsync sort of does the job on BSD (at too
high a price), as does O_SYNC on some Sytem V's.  Unfortunately, O_SYNC
is a bit too zealous, as it forces a write of both the data block and
the inode (for mod time).  The latter is often overkill, and might
better be a separate operation.  You don't always care if the mod time
is right, and not want to pay the cost of another real disk i/o.

I don't know if OS/2 does _the_right_thing_ in this regard, thogh it
seems to have some flavor of the feature.  Maybe it's like VMS:  no OS
cache means guaranteed writethrough :-)

-dB

"I don't care what you say, as long as you spell my name right."
{amdahl, cbosgd, mtxinu, ptsfa, sun}!rtech!daveb daveb@rtech.uucp