Re: "Transactions are bad, real bad" - discuss
Date: Mon, 12 May 2003 09:16:58 -0700
Paul Vernon wrote:
> "Costin Cozianu" <c_cozianu_at_hotmail.com> wrote in message
>>Paul Vernon wrote: >> >>>"Pablo Sanchez" <pablo_at_dev.null> wrote in message >>>news:Xns9373EB0EF7592pingottpingottbah_at_126.96.36.199... >>>[snip] >>> >>> >>>>My school of thought is that a DBMS shouldn't be the end-all >>>>solution. >>> >>> >>>I would be genuinely interested in knowing why you think that. Do you have
>>>deep reasons? Maybe you just want to protect the job creation scheme that
>>>the role of 'Applications Programmer' ???? >>> >> >>I'll give you another reason: intractability. Some problems are >>unsolvable in the database, or at least in the database as we understand >>databases now. >> >> >>Unless you want the database language to have the full programming power >>*and* expressivity as a full blown language - say LISP, and massage the >>whole echillada in the DBMS, which would be truely horrible.
> You make a good point Costin. My take is that we do want full expressive power
> in the DMBS, but that we also want to restrict that power and not make it
> available to untrusted users.
You want to isolate high-risk computations from the core RDBMS.
In real world computations fial, and sometimes they fail horribly. If for nothing else than for the necessity to integrate third party packages, it's always the thrid party and not your code you know :)
> What I'm thinking is that the language available on a DBMS is a nesting of
> subsequently more expressive languages.
> So there might be a basic relational core - without generalised transitive
> closure - that could safely be used by even malicious database users. From
> there, the more trusted you are, the more powerful syntax you are allowed -all
> the way up to full expressivity of say some functional language.
> I'm not sure how such a language would look, and I guess that logically it
> would actually be multiple languages and if only the relational subset was
> seen as the "DBMS language", then maybe I have agreed with your point, we
> don't necessarily want full expressivity in the core relational model...
> Paul Vernon
> Business Intelligence, IBM Global Services
So even if you give a more expressive language to power users to create user defined types and operators and what not, from a "software architecture" point of view, building all your important computations under the RDBMS umbrella and letting clients only do the UI rendering job ( at least this much I understood from your vision) is an absolute no go.
Costin Received on Mon May 12 2003 - 18:16:58 CEST