Posted by Tony Rogerson on 02/07/06 16:35

Hi Serge,

You are right, sorry Serge, not sure what I was thinking of when I stated
that, anyway, this link gives some more information on its current
performance http://www.microsoft.com/sql/prodinfo/compare/tpcc.mspx.

It will use parallelism from the perspective of the OS, Windows uses
threads, SQL Server allocates a number of 'worker threads' that connections
can use, so multiple connections will use multiple threads, one connection
will only use one CPU at a time, unless, the query runs using parallisation
which means that the query is broken up into streams and each stream
allocated a processor (via threads) and then executed in parallel and then
merged back at the end.

Query plans that use parallelism tend to need to do a lot of work so are
like you say best in a reporting / datawarehouse environment, assuming that
you haven't used Analysis Services to build your dimensional MOLAP database.

But, going back to my point, celko is completely wrong, SQL Server does do
parallelism and does have partitioning and has done for 3 versions of the
product now.

Tony.

--
Tony Rogerson
SQL Server MVP
http://sqlserverfaq.com - free video tutorials


"Serge Rielau" <srielau@ca.ibm.com> wrote in message
news:44rlpvF3kmf4U1@individual.net...
> Tony Rogerson wrote:
>> Partitioning works fine, in fact, it helped SQL Server to become the
>> first product to break then 1million TPS on the TPC benchmarks - google
>> it.
> Are you refering to TPC-C and TpmC? If so the first to break 1M was Oracle
> http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=103110401
> The only > 1M result I see from MS is this one:
> http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=105112801
>
> I do very much doubt that any vendor including SQL Server is using SMP
> paralelism in the TPC-C benchmark because the load is driven by the high
> number of concurrent users. SMP parallelism is mostly useful for low
> concurrency enviroments (such as warehouses (TPC-H)).
>
> Cheers
> Serge
> --
> Serge Rielau
> DB2 Solutions Development
> DB2 UDB for Linux, Unix, Windows
> IBM Toronto Lab

[Back to original message]