Re: What happened to Dali/Datablitz?

>
> Like TimesTen, and two or three other 'main memory database systems',
> they were shown in practice to be slower than commercial DBMS engines,
> and less reliable.
>
To shorten this discussion:
IMDB = In-memory DB
DBMS = Disk Based DB Management System

> A database is either too big to fit into memory, or it isn't. If it
> was too big for main memory (95% are) then these systems simply
> weren't useful at all.
Well, if it doesn't fit in memory you don't use an IMDB the way it's intended to be used. It's like using too big database caches in a DBMS and use the swap and say that's ok since it's already disk based.

>
> Beyond that, a database workload is either read-only, or read-write.
> When your database fits into main memory and your application involves
> write operations, you need transactions, which means disk operations.
> And that means main-memory DBMSs have no advantage anymore, even in
> theory. Twenty years of hand-polishing transaction management code
> makes commercial DBMS products superior.

I have used such superior systems and quite a few of them seem to need another 20 years of polishing.

>
> When your database fits into memory and your workload is read-
> intensive, the advantage of main-memory data structures over data
> structures intended for disk-and-memory is small. Having a rich
> feature, mature set (security, query optimization, good tooling) on
> the other hand becomes the deciding factor.
>

I work in a field where a DBMSs simply is no option unless you want to buy piles of HW; and to say mine and other high volume OLTP apps is only 5% of the market is great comedy.

> Of course you can find benchmarks where one of the main memory DBMSs
> outperforms one of the commercial DBMSs. It's just that there weren't
> that many customers who's problem was the one benchmarked.

I measure myself, I don't need biased quotes.

> Like TimesTen, and two or three other 'main memory database systems',
> they were shown in practice to be slower than commercial DBMS engines,
> and less reliable.

Done wasting my time on this:
:-) Received on Tue Jul 10 2007 - 08:33:41 CEST