[PVFS-developers] sc2003 BoF wrapup
Bob Arctor
curious@pb194.luban.sdi.tpnet.pl
Thu, 4 Dec 2003 11:32:58 +0100
> . We talked a bit about optimizations we can make for read-heavy and
> read-only applications (and it turns out there are quite a few of
> those applications). RobR went out on a limb and promised support
> for an "immutable" flag by SC2004 -- even if it meant doing a
> bunch of work a couple weeks before the show :>
while now there are just a few, terabyte-class ro devices open new possiblities to compilers and programmers.
let's take simple, and well known problem :
imagine problem of 2*2=4
while we do know it is 4 , computer have to calculate it. while 2bit * 2bit variable is simple,
multyplying two double long variables might take some time, and it may be faster to just ask
filesystem wheter there isn't file named like those two digits in mul/results/ directory.
ofcourse many modern cpu's use special devices to perform similiar task by finding base in 'quick'
ROM device, and then just multiply remaining bits,and adding to result found in ROM, but there are many more 'ready results' which filesystem
can 'memorize' and which should to be duplicated by caching mechanisms as close to computing system as possible.