380 likes | 543 Vues
ATLAS High Level Trigger. Introduction System Scalability Trigger Core Software Development Trigger Selection Algorithms Commissioning & Preparation for Cosmics & First Beam. ON-line. OFF-line. Level-1 Trigger 40 MHz Hardware (ASIC, FPGA) Massive parallel Architecture Pipelines.
E N D
ATLAS High Level Trigger Introduction System Scalability Trigger Core Software Development Trigger Selection Algorithms Commissioning & Preparation for Cosmics & First Beam
ON-line OFF-line Level-1 Trigger 40 MHz Hardware (ASIC, FPGA) Massive parallel Architecture Pipelines Rate (Hz) Event rate QED 8 10 Level-2 Trigger 100 kHz s/w PC farm Locale Reconstruction 6 10 Level-1 2 µs 4 10 1 sec Level-2 W,Z 2 10 Massstorage Top 10 ms Z * Reconstruction& Analyses 0 10 TIER0/1/2 Higgs Centers -2 10 Level-3 Trigger 1 kHz s/w PC farm Full Reconstruction -4 10 Offline Analyses sec 25ns 3µs ms year hour 10 10 10 -9 -6 -3 10 10 10 sec -0 3 6 Introduction
Introduction • ATLAS trigger comprises 3 levels • LVL1 • Custom electronics & ASICS, FPGAs • Max. time 2.5ms • Use of Calorimeter and Muon detector data • Reduce interaction rate to 75 kHz • LVL2 • Software trigger based on linux PC farm (~500 dual CPUs) • Mean processing time ~10 ms • Uses selected data from all detectors (Regions of Interest indicated by LVL1) • Reduces LVL1 rate to ~1 kHz • Event Filter • Software trigger based on linux PC farm (~1600 dual CPUs) • Mean processing time ~1s • Full event & calibration data available • Reduces LVL2 rate to ~100Hz • Note – large fraction of HLT processor cost deferred initial running with reduced computing capacity
40 MHz 1 PB/s FE Pipelines 75 kHz ROD ROD ROD RoI Read-Out Drivers 120 GB/s 120 GB/s RoI data = 1-2% Read-Out Links ROS ROB ROB ROB RoI requests ~ 10 ms LVL2 Read-Out Buffers ROIB L2SV RoI Builder L2 Supervisor Read-Out Sub-systems L2N L2P L2 N/work L2 Proc Unit ~2 kHz ~2+4 GB/s Lvl2 acc = ~2 kHz Dataflow Manager EB EBN Event Building N/work ~ sec Event Filter SFI Sub-Farm Input DFM ~4 GB/s Event Builder EFN Event Filter Processors Event Filter N/work EFP EFP EFP EFP EFacc = ~0.2 kHz SFO Sub-Farm Output ~ 200 Hz ~ 300 MB/s ATLAS Trigger & DAQ Architecture Trigger DAQ Calo MuTrCh Other detectors 40 MHz specialized h/w ASICsFPGA LV L1 D E T R/O 2.5 ms Lvl1 acc = 75 kHz H L T D A T A F L O W
LVL1 decision made with calorimeter data with coarse granularity and muon trigger chambers data. Buffering on detector LVL2 uses Region of Interest data (ca. 2%) with full granularity and combines information from all detectors; performs fast rejection. Buffering in ROBs EventFilter refines the selection, can perform event reconstruction at full granularity using latest alignment and calibration data. Buffering in EB & EF ATLAS Three Level Trigger Architecture 2.5 ms ~10 ms ~ sec.
LVL1 - Muons & Calorimetry Toroid Muon Trigger looking for coincidences in muon trigger chambers Calorimetry Trigger looking for e/g/t + jets • Various combinations of cluster sums and isolation criteria
ATLAS LVL1 Trigger pT, h, f information on up to 2 m candidates/sector(208 sectors in total) ET values (0.20.2)EM & HAD ET values (0.10.1)EM & HAD O(1M) RPC/TGC channels ~7000 calorimeter trigger towers Muon trigger Calorimeter trigger Muon Barrel Trigger Muon End-cap Trigger Pre-Processor (analogue ET) Jet / Energy-sum Processor Cluster Processor (e/g, t/h) Muon-CTP Interface (MUCTPI) Multiplicities of m for 6 pT thresholds Central Trigger Processor (CTP) Multiplicities of e/g, t/h, jet for 8 pT thresholds each; flags for SET, SETj, ETmiss over thresholds; multiplicity of fwd jets Timing, Trigger, Control (TTC) LVL1 Accept, clock, trigger-type to Front End systems, RODs, etc – RoI pointers
LVL1 triggers on high pT objects Calorimeter cells and muon chambers to find e/g/t-jet-m candidates above thresholds 2 2e RoI Mechanism • LVL2 uses Regions of Interest as identified by Level-1 • Local data reconstruction, analysis,and sub-detector matching of RoI data • The total amount of RoI data is minimal • ~2% of the Level-1 throughput but it has to be accessed at 75 kHz H →2e + 2
Physics Selection Strategy • ATLAS has an inclusive trigger strategy • LVL1 Trigger on individual signatures • EM cluster • Muon track • Jets • Total Energy • Missing Energy • LVL2 confirms & refines LVL1 signature • requires seeding of LVL2 with LVL1 result – i.e. RoI • Event Filter confirms & refines LVL2 signature & more complete event reconstruction • Possibility of seeding of Event Filter with LVL2 result • tags accepted events according to physics selection • Reject events early • Save resources • minimize data transfer • minimize required CPU power
ATLAS TDAQ Physical Layout Central Switches Events Built
System Scalability • Extended testing programme for system scalability testing • Dedicated testbed for dataflow performance & networking issues • Data Acquisition group • Large clusters worldwide for “node” scalability testing • Machine & run control • Start/end run cycling • Software distribution • Large scale configuration • Data Acquisition & Trigger groups • Trigger focus on Event Filter • Recent work • Use of LXSHARE cluster at CERN ~ 500 nodes and WESTGRID cluster in Canada (~840 nodes) • Plans • Use of 50-700+ nodes on LXSHARE this summer • http://atlas-tdaq-large-scale-tests.web.cern.ch
Summary of Recent Tests • Conclusions • Primary goal was system porting and debugging • Important bug in CORBA lib was found and fixed • Many others benefits obtained: • Experience in porting large-scale DAQ system • Many particular indications for weak points and possible improvements • General impression of run control transition times • LST @ CERN • June 6 – July 19 • Many things being tested / investigated / measured • We are ready following experience from WestGrid
System Scalability • Many hardware issues need attention • How to organize O(2000) PCs • racks, space, weight, heat & cooling, cabling • data I/O & networking • operating – booting, s/w installation, operational monitoring • dependency on ever evolving PC & CPU architectures and compilers, applicability of Moore’s Law • Remote farms • Possible Involvement • Longer term possibilities of LSTs at SLAC? • Software development & testing work in the Event Filter to include requirements from overall ATLAS monitoring and calibration • Work on the specification development, installation, maintenance & running of the EF
Trigger Core Software Development • Provides a coherent software framework for LVL2 and EF • Coherent data access methods • Re-use of some offline components where appropriate • Development platform ~common across trigger & offline • Facilitates online/offline comparisons & ease of development • Detailed collaboration with core offline development group as well as detector software development • Benefit from detailed expertise in each detector group • E.g. => in last year’s testbeam: detector monitoring software developed for use in offline was also used online in the EF • Considerable exchange of ideas & development • Performance & efficiency improvements done for the trigger now benefit offline some new offline functionality benefits the trigger • More specific dedicated development for LVL2
Event Filter HLTSSW Processing HLT Core Software Task 1..* Steering Monitoring Service HLT Algorithms L2PU Application Data Manager HLT Algorithms ROB Data Collector 1..* Event Data Model MetaData Service <<import>> <<import>> <<import>> <<import>> Package Athena/ Gaudi Event Data Model Reconstr. Algorithms StoreGate Interface Offline Reconstruction Offline Core Software Dependency HLT Event Selection Software HLT Selection Software • Framework ATHENA/GAUDI • Reuse some offline components • Common to Level-2 and EF HLT Data Flow Software ~Offline algorithms used in EF
Support for multiple threads Online Offline Data Flow ATHENA Environment L2PU athenaMT Steering Controller Steering Controller Link to algorithm libraries Algorithms Algorithms LVL2 Development Environment • HLT software development and testing in offline environment • Final “certification” procedure in Data Flow test-beds Offline support for Level-2 developers Multithreaded offline application AthenaMT Emulates complete L2PU environment No need to setup complex Data Flow systems As simple to run as a normal offline application: athenaMT <number of threads> <job-configuration> Coding guidelines for Lvl2 developers Development and Data Flow setup for Level-2
Trigger Core Software Development • Possible Involvement • Work & responsibility in specific s/w packages in the core s/w • Trigger configuration and algorithm control system • Trigger monitoring framework and strategy • Offline/online Software integration
Trigger Selection Algorithms • On-line event selection in the HLT based on algorithmic software tools running in LVL2 and EF farms, sequenced by HLT steering • LVL2 specialized algorithms, EF algorithms adapted from off-line • Important deployment in HLT test-beds to assess compliance with realistic on-line environment • Building on expertise and development inside detector communities • Calorimeters, Inner Detector, Muon Spectrometer • Studies of efficiency, rates, rejection factors, physics coverage organized around five main lines (“vertical slices”) coherently mapped to the Physics Combined Performance groups (see physics session) • Electrons and photons • Fundamental signatures for both precision measurements and discovery signals • Muons • Low- and High-PT objects, strategic also for B-physics programme • Jets / Taus / ETmiss • Models testing, new physics • b-tagging • Optimize physics coverage, add flexibility and redundancy to HLT selection starting from LVL2 • B-physics • Rich program of work with new strategies dependent on luminosity • Most recent talks on performance studies • http://agenda.cern.ch/fullAgenda.php?ida=a052747
Trigger Menus and Strategy • Extracting tiny signals out of huge backgrounds requires the HLT selection strategy to be robust, redundant and flexible • Selections are mostly inclusive, with as-low-as-possible pT thresholds for fundamental objects • The usage of software tools at both HLT levels allows detailed studies of the boundary between LVL2 and EF • Different paths leading at approximately thesame efficiency (electrons in the figure) • Example of flexibility and different selection sequences • Choice will depend on background conditions, detector knowledge, luminosity, … • The building of complete Trigger Menus evolves and complement the work done in the slices • Moving from single objects to complex topological signatures • Include issues of pre-scaled triggers, monitor triggers, etc • Optimize to environmental conditions • Commissioning the HLT selection will be an important step towards physics data taking • Needs to be ready for cosmic period • Implies modification to algorithms, new sequences
Trigger Selection • Possible Involvement • Work in trigger algorithm development and selection performance evaluation • Jet / tau / Etmiss area is in particular need of increased effort • Other areas would also benefit from new manpower and groups willing to take on new responsibility • Preparation/adaptation of sets of algorithms & selection procedures for use in cosmic running and in initial beam periods (single beams, very initial collisions etc)
Commissioning • Detailed planning for stepwise commissioning of the trigger system (LVL1 & HLT) is being prepared • Planning taking account of detector plans and triggering requirements for their commissioning • Planning in various phases with increasing levels of integration • Commissioning planning is broken in 4 broad phases: • Subsystem standalone commissioning • Integrate subsystems into full detector • Cosmic rays, recording data, analyze/understand, distribute to remote sites • Single beam, first collisions, increasing rates • Phases will overlap • TDAQ “pre-series” system
TDAQ Pre-series system • Fully functional, small scale, version of the complete HLT/DAQ system • Equivalent to a detector’s ‘module 0’ • Purpose and scope of the pre-series system: • Pre commissioning phase: • To validate the complete, integrated, HLT/DAQ functionality • To validate the infrastructure, needed by HLT/DAQ, at point-1. • Installed at point 1 (USA15 and SDX1) • Commissioning phase • To validate a component (e.g. a ROS) or a deliverable (e.g. a Level-2 rack) prior to its installation and commissioning • TDAQ post-commissioning development system. • Validate new components (e.g. their functionality when integrated into a fully functional system). • Validate new software elements or software releases before moving them to the experiment.
Pre-Series 5.5 USA15 SDX1 Partial ONLINE rack-TDAQ rack- 4 HLT PC(monitoring) 2 LE PC(control) 2 Central FileServers Partial EF rack-TDAQ rack- 12 HLT PCs Partial Superv’r rack-TDAQ rack- 3 HE PCs One ROS rack-TC rack+ horiz. cooling- 12 ROS 48 ROBINs One Full L2 rack-TDAQ rack- 30 HLT PCs RoIB rack-TC rack + horiz. cooling- 50% of RoIB Partial EFIO rack-TDAQ rack- 10 HE PC(6 SFI - 2 SFO - 2 DFM) One Switch rack-TDAQ rack- 128-port GEth for L2+EB ROS, L2, EFIO and EF racks : one Local File Servers, one or more Local Switches
Commissioning • Phase 1 commissioning will be completely defined after the experience with the pre-series • Parallelize commissioning work as much as possible • Use data taken during detector commissioning to test data unpacking tools • Develop special algorithms to test component units • Extend offline s/w testing procedures • Provide infrastructure to collect systematic information from trigger selection studies: • List of selection variables • Graphs of rate and efficiency variation • There is a strong coupling with the offline commissioning activities • Trigger commissioning extends well into data-taking • Need good coordination with physics groups • Treat the trigger as a single object to be commissioned (inc. LVL1) • Will need a clear strategy for the daily run meetings (data request) • It is clear that the “Extra Triggers (monitoring, calibration, etc…) will be much larger than the foreseen 10% during the first months of data-taking
Commissioning • Possible involvement • We would like to benefit from your experience in commissioning and running the BaBar experiment & elsewhere • Work in installing, developing and exploiting the pre-series system • Development of algorithms and procedures that allow to rapidly check the trigger performance with real data and monitor the overall HLT commissioning advancement • Responsibility in the more general trigger commissioning activities and in preparing the ATLAS trigger for cosmic tests and first beams in LHC • There is considerable lack of effort in this area and there is room for major involvement and responsibility
Summary • Outlined several areas within the ATLAS HLT system where members of the SLAC team could contribute and take responsibility • Spread of areas ranging from more technical software design and implementation to much more physics oriented work • Many interesting challenges ahead to lead ATLAS into data-taking and first physics • TDAQ Workshop in Mainz, Germany 10-14 October 2005 WELCOME !!!
ATLAS LVL1 Trigger 75(100) kHz 75(100) kHz 75(100) kHz 75(100) kHz LVL1 Accept 75(100) kHz
ZRPC 2 m-RoI reconstruction at LVL2 using mFast Z MDT m ZRPC 1 RPC station 2 (Pivot) BML DZ = (Z RPC 2 + Z RPC 1)/2 – ZMDT T RPC station 1 (Low Pt confirm) Z Muon Road
mFast latency is the CPU time taken by the algorithm without considering the data access/conversion time: the presence of Cavern Background does not increase the mFast processing time. The total latency shows timings made on the same event sample before and after optimizing the MDT data access. Optimized version: total data access time ~ 800 ms; data access takes the same cpu time of mFast; muFast Timing Measurements • Optimized coderun on • (Pentium III @ 2.3GHz). • Physics: single muon,pt=100 GeV • Cavern Background: High Lumi x 2
Stepwise HLT Selection • Selection takes place in steps • Rejection can happen at every step • Trigger Decision and Data Navigation is based on Trigger Elements • Algorithms use the result from previous steps (Seeding) using the Data Navigation and the Trigger Elements • The initial seeds for the LVL2 steps are the LVL1 RoIs Event Accepted e50i+e50i ? e50i e50i + isolation isolation Decision e50 e50 + elecId elecId EM50 EM50 + RoI RoI LVL1 Trigger Element
The Different Commissioning Phases (1) • HLT standalone commissioning • Units of racks (considered to be a unit to be commissioned) • A rack delivered from installation has: • Checked the power, cooling and network within and outside the rack • Operating system installed • Commissioning starts with the installation of the DAQ and offline software • Check internal Dataflow (preloaded data) • Monitoring tools • Offline software • Offline software distribution procedures • Automatic testing procedures • Testing algorithms
The Different Commissioning Phases (2) • Integrate subsystems into the full detector. • These operations that have a very strong coupling with the offline commissioning activities • First start with data unpacking algorithms • Monitoring infrastructure to check this step • Use any commissioning data taken by the detectors to debug this part of the system • Even if the data is corrupted, it might be very useful to test the robustness of the code • Current activities (or areas where we need to concentrate effort) • Extend the pool of data prep algorithms • Algorithms must be scrutinized and broken up in simpler testing units • Testing procedures for both offline selection software and interface to DAQ software are being strengthened and running in the nightly automatically • The goal is to arrive to a set of tests that almost guarantee further test-bed (or pre-series, etc) integration will succeed • Specify constraints and tests in the offline software before distribution • Software distribution
The Different Commissioning Phases (3) • The remaining phases correspond to commissioning while data is being taken and assumes: • Complete HLT Dataflow is working • The algorithms start selecting/rejecting events • The trigger work will focus more on demonstrating that an algorithm gives an Xx.Yy% selection efficiency with some rejection rate • This activities are very important: • Help to develop and tune the algorithms • Give us the building blocks to test the complete HLT chain • However, for commissioning, we need to be focused also in some other aspects • Have a centralized place where the complete set of parameters that algorithms use (will be inside the configuration in the future) are listed • Size of data request around the ROI • Set of selection cuts • For every “selection variable” we need the graph of variation in selection efficiency and rejection rate around the chosen optimal point (we are sure we will have to tune it with data) • Need to prepare a set of algorithms and methods that allow us to check the trigger performance with data: • Particles with known mass (selected only triggering in one of its decay products) • How many hours of data-taking do we need to know the selection efficiency within a 5% precision?