BIND Threading and Performance
foo at bar.baz.invalid
Tue Mar 8 01:29:23 UTC 2005
Diosko <diosko at hotmail.com> wrote:
> This is an another question to continue... I have BIND 9.2.4
> compiled as 64-bit aplication... well everything seems fine, but -
> is it realy for BIND advantage? It runs on Sun Netra T1, Solaris 9
It depends :) If you need > 4GB addressing in named it is an advantage.
If you need < 4GB addressing in named it may still be an advantage.
In the broadest handwaving terms, going to 64bit may slow things down
- variables of type "long" and pointer (in LP64 at least, although
Windows uses P64, go figure - I'm not sure if anyone uses ILP64 (Int
Long Pointer 64 bits)) are twice as large as they were before, which
means added pressure on the processor data caches. The increase in
cache miss rate decreases Instructions Per Cycle (IPC) (or increases
Cycles Per Instruction - CPI - if you prefer to count that way).
If you had plenty of cache left over to pressure it may just be a
There are some cases where compiling 64-bit still may be a win, but
not because something is 64-bit per se:
*) In the long ago and far away, compiling as a PA2.0W (W == Wide ==
64 bit) application got one the "new" shared library calling
conventions which were more efficient than the PA 1.1/1.0 conventions
(those being 32 bit). This benefit wasn't from being 64 bit except
that at the time a 64-bit PA-RISC binary had no legacy requirements so
it was possible to change the convention without having to worry about
*) In the not so long ago, there was at least one CPU where compiling
64-bit got you extra registers. This is the AMD64 stuff. Again, (and
IIRC) this isn't really a function of being a *64-bit* binary. It was
another case of there being no legacy with which to deal with AMD64
being a new binary type so adding registers was possible without
concerns about legacy requirements.
Process shall set you free from the need for rational thought.
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to raj in cup.hp.com but NOT BOTH...
More information about the bind-users