<html>
Dear colleagues,<br>
<br>
<x-tab> </x-tab>while
compression is still topical, I thought I'd share with you some results
of tests on compression vrs tiling which reminded me of something I'd
forgotten: the more tiles you have the less the gains in the first level
of compression (I assume everyone will use double compression to save on
archive space and remove the second level of compression only when
processing NeXus files). <br>
<br>
I generated NeXus files with different numbers of equally sized tiles
(SDS's). The size of the file after the second level of compression
depended very little on the first level of compression. I used a TOF
neutron single crystal Ruby data 80x80x150 unsigned short integers for
the tests.<br>
<br>
<x-tab> </x-tab><x-tab> </x-tab>File
Size in MB after first level of compression<br>
<br>
Number of Tiles
<x-tab> </x-tab>1<x-tab> </x-tab>2<x-tab> </x-tab>4<x-tab> </x-tab>8<x-tab> </x-tab><br>
LZW<x-tab> </x-tab><x-tab> </x-tab><x-tab> </x-tab>.257<x-tab> </x-tab>1.12<x-tab> </x-tab>1.51<x-tab> </x-tab>1.70<br>
SH<x-tab> </x-tab><x-tab> </x-tab><x-tab> </x-tab>.366<x-tab> </x-tab>.934<x-tab> </x-tab>1.05<x-tab> </x-tab>1.47<x-tab> </x-tab><br>
RLE<x-tab> </x-tab><x-tab> </x-tab><x-tab> </x-tab>1.03<x-tab> </x-tab>.938<x-tab> </x-tab>1.40<x-tab> </x-tab>1.65<x-tab> </x-tab><br>
no
comp'<x-tab> </x-tab><x-tab> </x-tab>1.87<x-tab> </x-tab>1.88<x-tab> </x-tab>1.89<x-tab> </x-tab>1.9<br>
<br>
<br>
<x-tab> </x-tab><x-tab> </x-tab>File
Size in MB after second level of compression<br>
<br>
Number of
Tiles<x-tab> </x-tab>1<x-tab> </x-tab>2<x-tab> </x-tab>4<x-tab> </x-tab>8<x-tab> </x-tab><br>
whole file
gzipped<x-tab> </x-tab>.213<x-tab> </x-tab>.361<x-tab> </x-tab>.359<x-tab> </x-tab>.357<br>
<br>
Despite these results there are very large advantages in using tiling (as
David Love first suggested to me) when handling very large files (some of
ours will be ~0.8GB) on PC's with limited memory and with a desire for
fast processing (faster IO to disk). <br>
<br>
<font color="#FF0000">One other thing related to my ignorance of
programming. I could not use the 1.3.1 API untill I had edited the header
file to select the appropriate calling convention for my PC (winNT
C++)<br>
# define
MANGLE(__arg)<x-tab> </x-tab><x-tab> </x-tab>CONCAT(__arg,_)<br>
Is this because I'm not using the API correctly or just becuase I didn't
read the installation note properly?<br>
<br>
</font><font color="#0000FF">And while I have your attention, my Fortner
Browser crasheswhen I try to browse files with many compressed SDS's. Has
anyone else found this?<br>
<br>
<br>
</font><br>
<div>Paul Langan</div>
<div>B-Division</div>
<div>MS M888</div>
<div>Los Alamos National Laboratory</div>
<div>Los Alamos NM87545</div>
<div>USA</div>
<br>
<div>Tel: 505 665 8125</div>
<div>Fax: 505 665 3024</div>
<div>email: langan_paul@lanl.gov</div>
<a href="http://lsdiv.lanl.gov/nsb/" EUDORA=AUTOURL>http://lsdiv.lanl.gov/nsb/</a>
</html>