Page History
| Warning | ||
|---|---|---|
| ||
The tool has been generalized and externalized. The last releases of source code, documentation and pending issues are now available at the new TRAC server. The specific issues and features for FERMI are in the documentation guide called TSkim for FERMI, which should be more complete than information below. |
Use of the skimmer on Noric machines
...
| No Format |
|---|
/afs/slac/g/glast/ground/GLAST_EXT/$CMTCONFIG/5.10.00/root /afs/slac/g/glast/ground/GLAST_EXT/$CMTCONFIG/5.14.00g/root /afs/slac/g/glast/ground/GLAST_EXT/$CMTCONFIG/5.16.00-gl1/root /afs/slac/g/glast/ground/GLAST_EXT/$CMTCONFIG/5.18.0000c-gl1/root |
For simple kinds of data, such as merit tuples, using a different recent ROOT release should be ok.
If $ROOTSYS is not defined, the skimmer will search for $GLAST_EXT/ROOT/v5.1014.0000g/root. If $GLAST_EXT is not defined, it will be set to /afs/slac/g/glast/ground/GLAST_EXT/$CMTCONFIG. If $CMTCONFIG is not defined, it will be set to rh9_gcc32.
...
| No Format |
|---|
/afs/slac/g/glast/ground/DataServer/v5r10v6r1/bin/skimmer |
Of course, it is more confortable to define an alias for it in your Unix account setup files, or add the DataServer bin directory to the Unix PATH :
| No Format |
|---|
sh> export PATH=<SKIMMER_DIR>/<release>/bin/skimmer:${PATH}
csh> setenv PATH <SKIMMER_DIR>/<release>/bin/skimmer:${PATH}
|
For what concerns the external tools, at runtime, the skimmer depends on :
...
The kinds of trees which can be skimmed is currently hardcoded, and especially targetted to GLAST needs. For each of those tree, we expect a top main branch. Actually, we distinguish two sets of trees : the ones which only rely on predefined ROOT types and have a single level of branches under the main one (the tuple-like trees), and the others. The latter typically requires the loading of a dedicated data description library before the concrete skimming takes place. The last release v5r10 v7r0 should understand the following types :
...
Worth to be noted, some different kind of trees can be grouped in common files, such as merit, pointing_history and jobinfo. When skimming, the result is split among several files, one for each kind of tree. Also, some kind of trees do not have branches for the run and the event ids, such as pointing_history and jobinfo. Those trees are never skimmed, rather always merged.
The subset of ROOT releases which can be combined with the skimmer
The skimmer top routine is a perl script which will execute a set of binaries. Those binaries are depending on ROOT libraries. The administrator of a given site, when installing a new skimmer release, must compile its binaries several times, for the different ROOT releases available at the site and supported by the skimmer. Later on, when one run the skimmer, the top perl routine will look at $ROOTSYS so to know if the user require a specific ROOT release, and it will use the corresponding set of binaries.
It is necessary to combine the skimmer with different ROOT release, because many kinds of data sets (such as mc) contains complex data structure and require the loading of the corresponding shared library (such as libmcRootData.so), and the later has been compiled with a specific ROOT release. The skimmer binaries which are used must be ones compiled with the same ROOT release, or the final behavior is undefined. That's the general case. Yet, for some simple kind of data sets (such as merit), typically tuple of predefined ROOT types, there is no need to load a shared library describing the data structure, and one should be able to skim with a recent ROOT release, the data generated with a different older ROOT release.
So, on any site, including noric cluster, the administrator has compiled the skimmer for a subset of possible ROOT releases, and the user must choose within this subset.