From mboxrd@z Thu Jan  1 00:00:00 1970
From: Vincent D Murphy <murphyv@student.cs.ucc.ie>
To: 9fans@cse.psu.edu
Message-ID: <20010208193128.A8122@student.cs.ucc.ie>
References: <200102081857.LAA20130@dns.irm.r9.fws.gov>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
User-Agent: Mutt/1.1.1i
In-Reply-To: <200102081857.LAA20130@dns.irm.r9.fws.gov>; from Mark_Otto@FWS.Gov on Thu, Feb 08, 2001 at 01:57:51PM -0500
Subject: [9fans] Re: Plan9 Relational Databases
Date: Thu,  8 Feb 2001 19:31:28 +0000
Topicbox-Message-UUID: 5e8c9bc0-eac9-11e9-9e20-41e7f4b1d025

= Mark Otto <Mark_Otto@FWS.Gov> [20010208 1859]:
> Do others have an idea of what a plan9 RDB FS should look like?

i think the real challenge is in providing ACID while keeping 9p simple,
clear, and general; the interface would follow.

i don't think defining the interface is the real challenge.  surely,
the structure and semantics of the fs could be devised the same way they
are for any 9p fileserver, (e.g. /net).  essentially the only argument
left to be resolved is how granular the fs should be; should files be
rows or entire tables (shooting from the hip, i would say the latter,
though taking efficient query evalutation into account may shift the
goalposts somewhat).

i wonder could joins be done with bind(2) though; that would be just
too cool.  :)

perhaps it's worth mentioning that the tux2 filesystem for linux promised
atomic writes using a method similar to that in netapp's woffle fs.
maybe that's somewhere to start.  http://innominate.org/~phillips/tux2