1

“Steering the ATLAS High Level Trigger”

COMUNE, G. (Michigan State University) GEORGE, S. (Royal Holloway, University of London)HALLER, J. (CERN)MORETTINI, P. (I.N.F.N. Genova)SCHIAVI, C. (University of Genova & I.N.F.N. Genova)STAMEN, R. (Institute of Physics, University of Mainz)TAPPROGGE, S. (Institute of Physics, University of Mainz)

Computing in High EnergyAnd Nuclear Physics13-17 February 2006

T.I.F.R. instituteMumbai, India

Thanks to the ATLAS collaboration

2

Outline

• LHC, ATLAS and ATLAS Trigger• Steering• Seeded and stepwise reconstruction• Configuration and operations• Persistency• Status and timing• Conclusions

3

LHC and ATLAS• LHC: 14 TeV CoM p-p collider (@ 40MHz)• High(Low) luminosity regimes: 1034 cm-2/s (2∙1033)

– High(Low) luminosity ~23(2) p-p collisions

4

ATLAS Trigger• Event size ≈ 1.5 MB • @ 40MHz => 60 TB/s!!• O(100MB/s) to tape• A three levels Trigger that

only uses “Regions of Interest” (of high pT activity) (RoI)

• Data access/preparation and reconstruction only done for a small fraction of the full event

• Full event building done only at much lower rate

5

Offline/Online portability• Complete off/online portability of HLT SW is

guaranteed by reusing offline software in the online system– Level-2 algorithms are custom written

• Use the same services and tools provided by the offline community

– Event Filter algorithms are imported directly from the offline reconstruction code

• adapted to the RoI guided reconstruction • Data converters provide the ability to run

unmodified code against online and offline event data

• Everything in the HLT SW and more specifically in the Steering is designed to allow complete offline/online symmetry– Trigger optimization, building of efficiency Vs rejection

curves, develop trigger strategies and physics menu tables, testing and timing can be done offline

6

High Level Trigger Selection SW

7

Steering• The Steering is responsible for:

– Unpacking the Level-1/Level-2 Result and creating the initial RoIs/seeds

– Sequencing the PESA algorithms according to static config and dynamic trigger conditions

– Accepting/rejecting events based on a static configuration and the dynamic event reconstruction outcome

– Provide data navigation and persistency tools to the algorithms for seeding and offline purposes

– Forming and handling event result• The Steering has been designed to cope

with the 10ms Level-2 latency

8

Seeded & Stepwise Reconstruction• Reconstruction is “initiated” at Level-2 by the Level-1 Regions

of Interests .• Event Filter starts off the Level2 result• The “outcome” of one “Trigger Algorithm” is the “seed” for the

subsequent algorithm• Reconstructed objects in a RoI are “linked” through

“navigational” links among them and back to the initial RoI • Reconstruction is broken down in “steps”

– Which algorithms run at what time is driven by a static configuration (“sequences” objects) matched against the dynamic event outcome (“satisfied” trigger conditions)

• Events can be rejected at any step– Reject/accept is driven by the static configuration

(“signatures/menus” objects) matched against the outcome

9

Reconstruction Chain

time

10

• The Trigger system is complex:– Hundreds of trigger items: single and multiple channels, combined triggers, pre-scaled

triggers• One must be able to change pre-scale during an LHC run

– In a predictable fashion• Must be able to study trigger signatures efficiencies

– Changing one pre-scale/threshold should not affect the whole system performance!!

• An “entangled” system is hard to operate and to understand• A coherent and disentagled configuration enforced by SW tools

– Allows to disentangle signatures and to easily pinpoint hot triggers– Changes to one signature do not affect the overall Trigger performance

Configuration and operations

11

Persistency• Trigger objects are recorded permanently

– For debuging, monitoring, calibration, tuning of physics analyses or selection strategies

• Two technologies are provided– Generic serializer (dictionary based)– Hand written serializer

• For offline:– Trigger optimization and tuning– Trigger algorithms and strategies development– Trigger aware physics analyses

• Completeness • For online:

– Level-2 => Event Filter seeding• Speed

– Event Filter => Storage• Completeness

• The Steering and the Persistency technologies have been developed to enable the development of online triggers in an offline environment using exactly the same SW

12

Status and Timing• The development of the Steering is well underway• Software used in the Test Beam, detector commissioning

and large scale MC data productions • Last missing features currently under development

– Topological triggers– Secondary RoIs – Pre-scaled Triggers

• The performance of existing software is well withing the Level-2 latency

100 signatures6 sequencesEvent Result formation

Pentium4 2.8 GHz

13

Conclusions• The specifications and high level

design of the Steering software is completed

• Implementation is advanced with positive timing results

• Work is ongoing to add remaining functionalities

• Focus is on the commissioning of the system and on providing the final tools to perform Trigger aware physics analyses