Re: [boost] [sorting] External Sorting

From: Sam Schetterer (samthecppman_at_[hidden])
Date: 2007-03-31 18:34:39

On 3/31/07, Richard Day <richardvday_at_[hidden]> wrote:
> Basically I would gather from this that a 10gb file for example would
> take 30gb of disk space to sort it ?
> That would appear to be what your suggesting here.
> And why 1/16 specifically ? Wouldnt it be better to use available memory
> (Free real not paged) if possible instead of specifically doing 1/16'th ?
> 10gb is more then most(Non-Server) machines have now that is true but
> 2-4gb is becoming more and more common now and of course servers may
> have way more then 10gb.
>
> Seems to me that if your going to do this much writing(2 X Original
> size) wouldnt it be simpler to just use a memory mapped file.
No, it is not 2x prignal size on disk. This illustration should help
o           o      o o   o          o         o     o  o o
 o   o o    o  o o o      o o    o   o   o   o o  o o
o  o o       o     o    o  o o   o     o   o  o     o
      o   o o   o   o    o   o   o o o   o o o
beomes (I leave out much data for simplicity)
       o       o       o       o       o
     o       o        o       o       o
   o       o         o       o       o
 o        o         o      o        o
o        o         o      o        o
which becomes
                                   ooooo
                           ooooo
                  ooooo
        ooooo
ooooo
all on one big file in memory. That is what I am trying to say.