floating point and Sun T2000 -- Michael Ebert 2006-05-05
What does "FP underpinnings for
scalability" mean? If you're concerned about performance, it is quite
likely that FP operations are faster than integer operations with the same
size on any hardware built in the last 10 years. Also I'd be surprised
at a CPU bigger than those used in cell phones that can't do FP ops in
hardware.
The famous Pentium FP bug, by the way,
occurred only under very special circumstances - it was discovered when
calculating pi to millions of digits. Even if you were running Oracle on
one of those, there would probably be a dozen other issues that should
worry you more.
Dr. Michael Ebert
DB2 & Oracle Database Administrator
aMaDEUS Data Processing
Erding / Munich, Germany
"Kevin Closson" <kevinc@polyserve.com>
Please respond to kevinc@polyserve.com
Sent by: oracle-l-bounce@freelists.org
05-05-06 19:22
To
<oracle-l@freelists.org>
cc
Subject
RE: floating point and Sun
T2000
>>>> >>>I'm testing a Sun T2000
Solaris 10 Oracle 10g and have
>>>read that if
>>>> >>>floating point calculations are used then
the box is
>>>essentially a
>>>> >>>1 cpu box.
>>>>
>>>> the database server does not do floating point calculations
...The question implied a scalability concern, right? My point is
that the engine does not have FP underpinnings for scalability. You
know, those things that happen, uh, like millions of times per second
like spinlocks, dba hashing, chain walks, etc. Of course I know about
things like BINARY_FLOAT and BINARY_DOUBLE datatypes, but I can't see
how these could ever be used to the point of rendering poor scalability.
Every time one of these data types is touched it is preceded by myriads
of integer ops. Tell me if I'm wrong by all means.
I know one thing for certain, we should all be glad the server doesn't
use floating point and that the NUMBER datatype is the way it is,
because that nasty little floating point bug in the pentium processor
back in the late 90s would have been a REAL pill to
swallow. Come to think of it, these new 10g datatypes would have
been
really freaky on those buggy pentiums :-)
