PartialEventBuilding

Introduction

Unlike physics events, calibration events for specific ATLAS subdetectors do not require the full event information, but only a fraction of the nominal 1.5~MB per event. In the ATLAS TDAQ system new features have been implemented which handle efficiently the assembly and logging of such calibration events based on a list of selected ROB or subdetector identifiers. These functionality allows an increased rate of calibration events within the allocated bandwidth and storage space.

Events for subdetector calibration purposes are selected by HLT algorithms, which create a list of ROB or subdetector identifiers and flag the event as a calibration event. The list of identifiers can be filled either in a static or a dynamic manner, based on the RoI information. The list is then passed to the SFI, which pulls the requested data fragments from the specified ROBs and assembles the calibration event. This method is a special case of the full event building and is referred to as "partial event building". It is only applied if the event does not fulfill any of the L2 physics selection criteria, but only a L2 calibration criterion.

The benefit of using partial event building can be illustrated by this expression:

Total event builder bandwidth = L2 accept * event size

The total input/output bandwidth for the event builder is about 5600MB/s and is fixed by the number of available event builder nodes (so-called SFIs) and the network. Thus, by reducing the event size a higher L2 accept rate can be achieved, while keeping the total bandwidth constant. The partial event building functionality is implemented in the SFI (Link to CVS) package of the TDAQ release. The algorithm that populates a list of ROB identifiers or subdetector identifiers is implemented in the athena package (TrigDetCalib). This algorithm allows to choose the size of the region of interest (RoI) used for the partial event building and the detector type can be configured for each calibration stream separately. Depending on the calibration requirements the list of ROBs or subdetector identifiers can be filled at L2 or filled/updated at the EF. In case the list of ROBs/subdetector is filled/updated at the EF, a partial copy of the full event which contains only the requested ROBs/subdetectors is created. This partial copy is obtained by using the so-called "event stripping" method and can either be called at the EFD or at the SFO. The event stripping is executed at the EFD for pure calibration events (i.e. they are rejected by all physics triggers) that receive an updated list of ROBs/subdetectors at the EF. The event stripping at the SFO is called for events which are accepted by calibration and by physics triggers. The stripped copy of the event is then send to the corresponding calibration stream.

The diagram below illustrates the data flow of calibration events.

calibrationdataflow.png

The ATLAS TDAQ functionalities described above are currently being used by the liquid argon (LAr) calorimeter group, the inner detector (ID) group and the tile calorimeter group. Details on the requirements of the different subdetector calibration streams and on the status and test results of the calibration data taking with partial event building and event stripping are given below.

Liquid argon calibration

For the liquid argon (LAr) calorimeter the needed calibration events are high-purity photons. Currently two trigger items are being used to collect calibration data: g10 and g20i, which refers to photons with a threshold of 10 GeV and 20 GeV, respectively. The "i" means that isolation criteria are applied. Such events get also selected from physics triggers, and thus get fully built before sent to the EF. The list of ROBs is then used at the SFO/EFD, in order to write a stripped copy of the event into the appropriate calibration stream.

Inner detector alignment

In order to provide frequent updates on the alignment constants, the inner detector group has requested to log such events at a rate of about 50~Hz. Assuming the nominal ATLAS full event size of about 1.6MB, this would result in 80MB/s. This needs to be compared to the expected total event rate after the EF of about 200Hz, which means 320MB/s. Without partial event building this means that the ID alignment would use up 25% of the available output bandwidth. This is obviously not possible and motivates the need for the partial event building.

For the inner detector alignment, events with isolated tracks are needed, and the associated RoI is used to fill the required list of ROBs. For the selection of the tracks the tau sequence is run first, selecting HA6 and HA9i L1 triggers. At Level-2 it selects single isolated tracks. The associated list of ROBs from the inner detector is passed from Level-2 to the event builder. The desired output rate at L2 is of the order of 50 Hz. For this L2 output rate and using a test luminosity of 10e31, the prescales for the L2 objects was optimized at 40 for trk9i_calib and 1 for trk16i_calib.

The principle of this calibration item is illustrated below:

ID_calib.png

offcenter beam collisions

Use beam collisions that are shifted by +/- 37.5 cm from the nominal interaction point to select isolated tracks.

Tile calorimeter calibration

The tile calorimeter uses a dedicated calibration system, which illuminates the full subdetector by either injecting laser pulses, discharging calibration capacitors into the read-out electronics or by using a Cesium source. Since the full tile calorimeter is illuminated by its calibration system, a static list of identifiers which contains all configured tile read out channels is used for the partial event building. A first test of these calibration triggers using partial event building was done in July 2008. The plots below show some distributions recorded in the SFI during Run 78852 on Jul 29 2008. Shown is the event size of the accepted events and the number of ROBs used for partial event building. The mean value of the event size distribution is at around 230kB and corresponds to the tile calibration event. Note the overflows in this histogram which correspond to fully built events and contain also the LAr detector information. For this run the LAr was using 32 samplings which explains the large event size beyond the limits of the histogram. The number of ROBs has a single peak at 65 which corresponds to all the 64 tile ROBs plus a special tile ROB used only for calibration runs.

tileCalibRun78852_evtSizeZoom.gif  tileCalibRun78852_nROBs.gif

Cosmic Runs

Run 91400

LVL2 online histogram for LArCalib_g10

CosmicLArCalib_nrobs.gif CosmicLArCalib_nrobsVsdet.gif CosmicLArCalib_ttemrobs.gif CosmicLArCalib_ttemrobs_zoom.gif CosmicLArCalib_tthecrobs.gif CosmicLArCalib_pixrobs.gif CosmicLArCalib_sctrobs.gif CosmicLArCalib_trtrobs.gif

To Do

new SFI histograms

event size Vs stream tag: The current SFI histogram with the event size does has only limited information. To tell that partial event building is working ok we need a histograms which shows the event size as a function of the stream tag.

Payload rate: A histogram showing the payload rate in MB/s is already implemented. However in the test runs this histogram is empty (has only zero bin entries). Currently investigating this.

Off beam center collisions

I am preparing an HLT algorithm for offline beam center collisions, but is not sufficiently tested yet.

Muon Calibration

Run in 24hr TDAQ period (scheduled August 3)

Test SFO/EFD stripping with TDAQ-02-00-00

The SFO stripping has only been tested so far (as of Dec. 3 2008) on the preseries as a private patch applied against TDAQ-01-09-01.

-- IgnacioAracena - 17 Jul 2008

Topic attachments
I Attachment History Action Size Date Who Comment
GIFgif CosmicLArCalib_nrobs.gif r1 manage 7.5 K 2008-10-16 - 16:52 IgnacioAracena  
GIFgif CosmicLArCalib_nrobsVsdet.gif r1 manage 14.2 K 2008-10-16 - 16:53 IgnacioAracena  
GIFgif CosmicLArCalib_pixrobs.gif r1 manage 31.0 K 2008-10-16 - 16:53 IgnacioAracena  
GIFgif CosmicLArCalib_sctrobs.gif r1 manage 22.9 K 2008-10-16 - 16:54 IgnacioAracena  
GIFgif CosmicLArCalib_trtrobs.gif r1 manage 34.0 K 2008-10-16 - 16:54 IgnacioAracena  
GIFgif CosmicLArCalib_ttemrobs.gif r1 manage 35.9 K 2008-10-16 - 16:55 IgnacioAracena  
GIFgif CosmicLArCalib_ttemrobs_zoom.gif r1 manage 24.8 K 2008-10-16 - 16:55 IgnacioAracena  
GIFgif CosmicLArCalib_tthecrobs.gif r1 manage 13.8 K 2008-10-16 - 16:55 IgnacioAracena  
GIFgif calibrationdataflow.gif r4 r3 r2 r1 manage 8.0 K 2008-12-04 - 04:37 IgnacioAracena  
PNGpng calibrationdataflow.png r7 r6 r5 r4 r3 manage 12.6 K 2008-12-04 - 04:38 IgnacioAracena  
GIFgif evtsize_sfi_ttbarevts.gif r1 manage 7.2 K 2008-07-31 - 16:30 IgnacioAracena  
GIFgif nrobs_sfi_ttbarevts.gif r1 manage 8.9 K 2008-07-31 - 16:30 IgnacioAracena  
GIFgif tileCalibRun78852_evtSize.gif r1 manage 6.6 K 2008-07-31 - 16:14 IgnacioAracena  
GIFgif tileCalibRun78852_evtSizeZoom.gif r1 manage 6.5 K 2008-07-31 - 16:13 IgnacioAracena  
GIFgif tileCalibRun78852_nROBs.gif r1 manage 8.3 K 2008-07-31 - 16:14 IgnacioAracena  
Edit | Attach | Watch | Print version | History: r8 < r7 < r6 < r5 < r4 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r8 - 2008-12-04 - IgnacioAracena
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    Sandbox All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2020 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback