Re: 9i server performance, 5 million hits per second?

S.Llabres doodled thusly:

>It is correct that the feature set is small in comparison to an Oracle
>RDBMS.
Exactly.

>But you cannot compare such an DB to "tables in memory" or an 3GL
>implementation (and they are real databases in the sense of a definition
>of an RDBMS).

No they most definitely are not. And they are much nearer to "tables in memory" than they are to a RDBMS.

>You would not have transactions, ANSI SQL, relational DB model
>and several other features with "tables" or a C/Pascal/whatever.

Transactions are completely pointless for a memory-only system. Other than helping to define a lock duration boundary. Which can be achieved in a memory-only system, much more efficiently, WITHOUT a transaction model. Or else queueing and conflict handling in Operating Systems would never have existed. And we'd still be running single processing stream systems.

>
>It is too correct that there are just memory lookups via a B-tree
>or T-tree. But whats wrong with that?

Absolutely nothing. Quite valid concept. Just don't call it a RDBMS or even a DBMS. It's not, it will NEVER be. No matter how much marketing hype one weaves around it.

>
>A lot cheaper depends. There are open source implementations available
>if you do not need the commercial support.

Why on Earth would I want to add a RDBMS framework library to ANY code to do a B-tree lookup? The darn library to do that has been public domain for eons!!! And I could even write my own: it was the subject of my C-programming exam, ages ago.

>The question was "5 million hits per second". That is some powers above
>that what an RDBMS can deliver today.

Nope. It's not even comparable. We are talking two completely different universes.

>I thought that it could interests some here (as the question was surely
>not for an real system) that there are applications with very high
>transaction requirements (not 5e6 though), even if do not think that
>these systems will replace the currently available RDBMS systems.
>

I'm aware of these systems through the OODB Usenet groups. Simple fact: they are not databases.

Efficient for their stated purpose, sure.

Although I believe they could be an order of magnitude faster. If they shed the "me-too" attitude and overhead of calling themselves databases and simply optimize their B-tree and T-tree scans, then they would be a lot faster.
Of course, the "glitz" of calling themselves what they cannot be would be gone.

Cheers
Nuno Souto
nsouto_at_optushome.com.au.nospam Received on Fri Feb 01 2002 - 07:26:04 CST