Re: Microsoft destroys TPC-C records!

First of all, I am not a big believer in benchmarks ... too much of a leapfrog ... 6 months later, all the positions will be changed, so what does this prove? A db decision must be made based on the answer to the question ... who do you want to do business with, long-term. Perhaps looking at the vendor/db that has been in the TPC top ten cumulatively and most consistently.

But let me address the use of the terminology "shared nothing". Shared nothing does not mean that the database is not being shared across nodes ... on the contrary. It means that the hardware is constructed/configured such that the memory/disk of each node is not shared. What does this mean for the database? It means that the database must still find a way of "sharing" the data across nodes. IBM does this via function shipping of SQL across nodes, determination of the result set and return of one unified result set to the requesting user. I don't know how Microsoft does it, but the term shared nothing is pretty much industry parlance for partitioning of data across nodes, and somehow sharing the partitions in some way, shape, or form without having the nodes actually share the same RAM or disk.

Ivana Humpalot wrote:

> During W2K launch Bill Gates said they are using a "shared nothing"
> approach. They are not even sharing the database, as far as I can
> tell. In other words each machine in the cluster is running its own
> copy of SQLServer and using a completely independent database. In
> other words each machines in the cluster is querying and updating a
> completely independent database.
>
> Can anyone confirm this is the case? And if true, does this have any
> significance in real life?
>
> "X X" <X_at_X.X> wrote in message news:38ace385$1$8288_at_readdo01.news...
> > Read 'em and weep.
> >
> > http://www.tpc.org/new_result/ttperf.idc
> >
> >
> >
> >
> >
> >
Received on Sun Feb 20 2000 - 14:42:40 CST