| Platform: | Linux Red Hat 7.3, in a 1.7Ghz AMD processor PC DB2 requires some kernel modifications (kernel capacity expansions) |
| Version: | V 7.1 |
| Data-sets: |
USNOA2, restricted to 3 million entries GCS2, 275000 entries The same as those used in the previous MySQL and Postgress evaluations. |
| Installation: | Instructions were reasonably good, although some omissions caused to repeat the installation. |
| Time to install: | a few hours |
| Operation mode: |
The test database account was used during the test. command line was mostly used, although a java based GUI is quite useful to do some checkup. |
| Language: |
SQL (but not the same flavour as MySQL or Postgress or Oracle) plus IBM's
own set of commands. There is also a Perl interface which was not tested but could be used as a CGI script or for Web Services. |
| Loading time: |
USNO: 30 seconds GSC: < 5 seconds |
| Index creation: |
USNO: 50 seconds. , disk usage: ~50Mbytes GSC: 5 seconds |
| Cone search: |
The main test was to cross correlate GSC with USNO to a certain limit.
The PCODE method suggested by C.Page was used. Time to perform the cross correlation only using Pcode : 74 seconds |
| Other queries: |
select count(*) from usno where rmag < 10: 5 sec
select avg(bmag) from usno where bmag < 13; 5 sec select stddev(bmag) from usno: 9 sec |
| Performance issues: |
An index was created on PCODE to make the search faster, but after
having to rebuild the system (omitting by accident to do the index),
DB2's performance was at the same level as with the indexation on PCODE done. In a second try, three columns were added to each database: the cosine directors, i.e., a cartesian representation of the sky position. Distance in the sky is the arc-cosine of the dot product between the cosine directors of the two points in question. This is done to avoid computing trigonometric functions over and over in data on both sides of the cross correlation. Not using PCODE and no indexation in any of the other variables suggested that the same cross correlation would take about 10 days (running continuously).
|
![]() |
Click here for the AstroGrid Service Web |
This is the AstroGrid Development Wiki |
|