[prev in list] [next in list] [prev in thread] [next in thread] 

List:       beowulf
Subject:    Compression/packing routines for double precision data
From:       Emil Briggs briggs () tick ! physics ! ncsu ! edu
Date:       1999-10-20 23:16:08
[Download RAW message or body]

>
>[ Packs no more bits than necessary to represent the real accuracy of
>  the data involved, instead of just sending 8 bytes ]
> 
>Yep - the meteorological world uses such a scheme for about a decade,
>now.  A good GRIB (GRIdded Binary) data converter includes an API which
>allows you to pass the accuracy with which to pack a field of data.
> 
>So while you need to know surface pressure with 24 bits of accuracy
>(mantissa only), relative humidity can be packed in about 6 bits.
> 
>However, as far as I know it is not used inside distributed memory
>versions of models to communicate data (mostly because the amount of
>communicated data is small anyway).                                                    
Interesting -- I guess it's being used there for storage reasons
instead of network bandwidth. In reference to Greg's post the real 
place for this is inside the message passing libraries which makes
it easy to experiment with and apply.

Emil

-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to beowulf-request@beowulf.org

[prev in list] [next in list] [prev in thread] [next in thread]