new version of QE checked into CVS

[Henrik Nordberg <hnordberg@lbl.gov>]

Hello Everybody,

I have checked in a new version of the Query Estimator into CSV.
The two biggest changes are better query performance and a higher
degree of portability. Scalability hasn't been addressed yet.

All programs (except mkbmp) under CVS/QE compile with CC 4.2, CC 5.0 
and egcs 1.1.2. This means QE runs on Linux. mkbmp doesn't compile
with CC 5.0. This is due to an error in <vector> that comes with CC 5.
(Or so it seems.)

It has been tested with the following configurations:

rmds03: CC 4.2 and 5.0
gizmo.lbl.gov: CC 4.2 and 5.0
Redhat Linux 6 (pcroot2.cern.ch): egcs 1.1.2

starsu00 could not be used because of licensing problems with both
CC 4.2 and 5.0.

The bitmap index is now faster than before. Query execution is CPU
bound. For queries with short ranges compared to the full range of
the attribute, the bitmap lookup is very fast. This is true also for
queries with long range compared to the total range, since for those
queries the complement to the query is calculated and the resulting
vector is then flip()ped.

The makefiles are cleaner now, but still not as clean as I'd like
them. There is still a problem with adding new machines to compile on.
In the future it would be nice to be able to use environment variables
for the paths to CORBA and STL (where it applies).

I will now start implementing the scalability changes we agreed on at
the Argonne meeting, along with the CORBA interface to the Index Feeder.

Cheers, Henrik
_________________________________________
Henrik Nordberg       <hnordberg@lbl.gov>
Scientific Data Management Research Group
Lawrence Berkeley National Laboratory