Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site osiris.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!gamma!epsilon!zeta!sabre!petrus!bellcore!decvax!genrad!panda!talcott!harvard!seismo!umcp-cs!aplvax!osiris!eric
From: eric@osiris.UUCP (Eric Bergan)
Newsgroups: net.database
Subject: Re: UNIX dbms
Message-ID: <493@osiris.UUCP>
Date: Thu, 29-Aug-85 15:43:10 EDT
Article-I.D.: osiris.493
Posted: Thu Aug 29 15:43:10 1985
Date-Received: Sun, 1-Sep-85 12:14:19 EDT
Distribution: net
Organization: Johns Hopkins Hospital
Lines: 48


	Time to add my two cents worth.

	A couple of years ago, I did a fairly extensive comparison of
Unify and Ingres (RTI version). There seemed to be some major failings
in Unify. The locking for concurrency control was insufficient, to do it
correctly required the applications programmer to determine what needed to
be locked. And even then, deadlock resolution was by timeout, not a more active
algorithm to detect deadlock.

	At the time of the comparison, Unify had not implemented any kind
of reliable journalling facility. This is extremely important in production
applications. I do not know if they have remedied this by now.

	In terms of performance, Unify was faster on simple things (selects,
appends, replaces), but much slower on more complex tasks, such as joins,
aggregates, and projects. While it was usually twice as fast on the simple
things, it was often orders of magnitude slower on the more complex
transactions. This was due to the lack of a query optimizer to try and 
determine the most efficient query tree.

	Also, while Unify does provide a query language, it does not
allow it to be embedded in programs. For the application developer, Unify
provides function calls that implement the access methods. But note: there
is no simple call to do a join, or an aggregate.

	As I have said, this review was done a couple of years ago, so
things may have changed. I also have not looked at Informix to see how
it handles these issues. Oracle, to the best of my knowledge, does provide
these more advanced features, but we did not do any performance testing
between Ingres and Oracle. The sales fliers for Mistress 32 also indicate
that it provides these more advanced features, but I can not personally
confirm or deny that claim.

	Basically, the bottom line again depends on what you are going to
do. Unify is probably better in smaller environments, or for embedding in
products that will not require much maintenance and do not support much
concurrent access to data. Ingres is a better choice when building larger
products, where data integrity is a bigger concern, and where the cpu
and disk horsepower are more likely to be there to support it.




-- 

					eric
					...!seismo!umcp-cs!aplvax!osiris!eric