LHCb data reduction (stripping)

The LHCb computing model foresees that not all events recorded by the data acquisition are made available for physics analysis. After an initial reconstruction of all events, an event selection step (stripping) is executed - this is an analysis job that selects events based on "stripping sekections" provided by the physics groups. Only events passing one or more of these selections are made available for further analysis.

Stripping framework

The stripping framework is a framework based on DaVinci within which the stripping selections are defined. More details are available here

Stripping workflow

The sequence of steps needed to produce a stripped DST ("stripping workflow") is shown in the following picture:

MC09 stripping workflow

What follows describes the workflow used to strip MC09 data, it refers to the picture above

In MC09, the RAW data was produced by Boole v18r1 and the RDST data by Brunel v34r7. In MC production, these datasets are not stored in separate files, they are combined in a single DST file which is the output of the Brunel production step.

Step 1: DaVinci (stripping)

This step reads the RDST from the MC09 production (taken from the MC09 DST files), it runs the default stripping and produces a “full” event tag collection (FETC): it stores the result of the stripping selection for each event on the RDST.

Program version:	DaVinci v24r2p2
Options	AppConfig/v3r4/options/DaVinci/DVStrippingETC-MC09.py
Database:	SQLDDDB v5r11
Database tags:	SIMCOND MC09-20090402-vc-md100
	DDDB MC09-20090602

The output is an FETC. This output is not stored permanently; it is only used as input to Step 2

Step 2: Brunel

This step runs the reconstruction to produce a DST containing all the events selected by Step 1. The inputs are the FETC produced by Step 1 and the RAW from the MC09 production (taken from the MC09 DST files), the output is a DST containing the OR of all the stripping selections. There are two possibilities for the version of Brunel:

1. The same version used in the original MC09 production. This guarantees consistency between the stripping done in Step 1, and that to be done in Step 3.
2. A more recent version, using the same version of REC as the version of DaVinci to be used in Step 3. This guarantees consistency of the reconstruction (in particular track pattern recognition) between Steps 2 and 3, but means that some of the events selected in Step 1 will not be re-selected in Step 3. Conversely, any events that would have been selected by step 3 but were not selected by step 1 are lost. Note that there isn’t yet a DaVinci version using the same REC version as the latest version of Brunel (v35r6) , only v35r5. Neither has undergone the extensive validation of the MC09 version.

Open questions

• Which of options 1. and 2. is preferred by the physics groups? The computing group prefers Option 1, since it has already been commissioned and is better tested. Note that option 1 is a pure computing exercise in the case of MC data: since the input is an MC DST and the output is also an MC DST produced by an identical programme, Steps 1 and 2 could be skipped in a future stripping of MC data that uses option 1.

	Option 1	Option 2
Program version:	Brunel v34r7	Brunel v35r5
Options	AppConfig/v2r9p1/options/Brunel/MC09-Stripping.py	AppConfig/v3r4/options/Brunel/MC09-Stripping.py
Database:	SQLDDDB v5r11
Database tags:	SIMCOND MC09-20090402-vc-md100
	DDDB MC09-20090602

The output is a DST. This output is not stored permanently; it is used as input to Step 3.

Step 3 DaVinci (stripping)

This step reruns the stripping selections taking as input the DST produced in step 2, and produces several DSTs, one per stripping stream. The stripping result is stored as a DST object. It re-runs the L0 and replaces on the RawEvent the L0 decision RawBank produced by Boole. It also runs HLT1 and stores the result as an additional RawBank on the RawEvent.

Program version:	DaVinci v24r2p2
Options	AppConfig/v3r4/options/DaVinci/DVStrippingDST-MC09.py
Database:	SQLDDDB v5r11
Database tags:	SIMCOND MC09-20090402-vc-md100
	DDDB MC09-20090602

Open questions

• Is this the right DaVinci version, or should we use v24r2p3? DaVinci v24r3 has some firther bug fixes and could be ready in time.
• Do the options above re-run L0, run HLT1, and store the results in RawBank as required? No.
• Are the L0 and HLT1 that are run using a correct TCK?
• Does the information in the RawBanks contain all that is needed for people interested in TIS/TOS studies?
• Is the stripping result stored on the DST?
• Are all B candidates and intermediate resonances stored on the DST? Yes.
• What happens when different selections in a stream select the same B candidates? It gets copied twice, since the output locations are different for each selection.
• Juan mentioned the possibility to save other information (eg. PV refit). Is any of this already in? (Not necessary for this stripping.) It the ReFitPVs property of the relevant DVAlgorithms is set to true, this gets done automatically. If not, some options have ot be set up.

The outputs of this step are one DST per stripping stream.

More open questions

• It is quite possible that some of the output DSTs will contain zero selected events, resulting in empty or non-existent output files. Is this a problem for DIRAC?
• Are the names of the streams OK and interfaced to the book-keeping?

Step 4: Merger

This is a technical step, does not require any specific application, just gaudirun.py with standard DST merging options. For each stripping stream it combines the (small) DSTs of individual stripping jobs into a larger DST whose file size is optimized for storage.

Step 5: Tagger

This is a DaVinci job that, for each stripping stream, reads the merged DST and uses the stripping result stored on the DST to produce a “selection” ETC (SETC). This SETC contains the stripping result for each event on the stripped DST, and can be used by analysis jobs to select only events passing a given stripping selection in a given stripping stream.

Open questions

• Which DaVinci version should be used? v24r2p2 or v24r2p3 (preference for the latter, but not a show-stopper if it isn't released in time)
• Which are the options for doing this, and are they released in AppConfig? New options need to include the necessary L0 and Hlt options.

-- MarcoCattaneo - 2009-09-29