RE: floating point and Sun T2000

From: Kevin Closson <kevinc_at_polyserve.com>
Date: Fri, 5 May 2006 11:36:03 -0700
Message-ID: <5D2570CAFC98974F9B6A759D1C74BAD06FEA17@ex2.ms.polyserve.com>

>>>

>>>The only time it's possible to have millions of floating
>>>points operations per second is while a nice
>>>SUM(amount_sold) on the large fact table is running.

Is amount_sold a 10g BINARY_FLOAT or BINARY_DOUBLE? If not that would not be fpu and the number of integer ops between each amount_sold++ far far far outweighs the cost of using scaled integer math. If the datatype is not BINARY_FLOAT or BINARY_DOUBLE, contrast the summation results from that SQL aggregation to the result you get if you pull that data out and cast it into a C floating point type and sum it there. Of course, amount_sold would unlikely require 38 digits of precision so the result should be the same....unless the dollar keeps dropping I guess.

BTW, I'm not sure the new BINARY_FLOAT datatype actually guarantees it will be mashed with the FPU. The docs only say it _can_ be so. Knowing Oracle, there is likely a port-level action to make it so.

--
http://www.freelists.org/webpage/oracle-l

Received on Fri May 05 2006 - 13:36:03 CDT