220 likes | 357 Views
IceTray A Software Framework for IceCube. Computing in High Energy Physics Interlaken, Switzerland. The IceCube Collaboration. University of Alabama University of California, Berkeley Clark – Atlanta University University of Delaware Institute for Advanced Study, Princeton
E N D
IceTrayA Software Framework for IceCube Computing in High Energy Physics Interlaken, Switzerland
The IceCube Collaboration University of Alabama University of California, Berkeley Clark – Atlanta University University of Delaware Institute for Advanced Study, Princeton University of Kansas Lawrence Berkeley National Laboratory University of Maryland Pennsylvania State University Southern University and A&M University of Wisconsin, Madison University of Wisconsin, River Falls DESY, Zeuthen Johannes Gutenberg-Universität Mainz BUGH Wuppertal Vrije Universiteit Brussel Université Libre de Bruxelles Université de Mons – Hainaut Stockholms Universitet Uppsala Universitet Imperial College, London University of Oxford Chiba University University of Canterbury, Christchurch Universiteit Utrecht Universidad Simón Bolívar, Caracas Amundsen-Scott Station, South Pole CHEP 2004
Atmospheric Muons & Neutrinos • “Atmospheric muons” from cosmic ray showers, penetrating to the detector from above • “Atmospheric neutrinos” from the same air showers, forming a diffuse background and calibration beam • Astrophysical neutrinos: interesting signal Astrophysical n cosmic ray Atm. n Atm. m cosmic ray CHEP 2004
Optical sensor • A neutrino interacts with a nucleus in the ice • A relativistic lepton is produced in the collision • Optical sensors map the lepton’s Cherenkov light • Timing, photon counting give direction & energy Muon Cherenkov light cone
The IceCube Detector IceTop 1 km3 instrumented volume: 1 Gton of ice 4800 digital optical modules (DOMs) in the ice on 80 strings AMANDA will be enclosed within the array IceTop air shower array on the surface above the detector (80 stations, 320 DOMs) Geometry optimized for TeV—EeV neutrinos South Pole IceCube AMANDA 1400 m 2400 m CHEP 2004
Muon Tracks CHEP 2004
Electron and Tau Events Electron: All energy deposited in a short cascade • Spherical diffusion Tau: “Double Bang” • nN interaction • t lepton decay • Separated at very high energies CHEP 2004
Software Framework • Want advantages of a modern framework • Modularity (better maintainability!) • Easy to reuse code, extend applications • Allows user/developers to focus on specific task • Reliability (core functions in common, well-exercised code) • Common style, shorter learning curve • What framework? • Some specific requirements similar to colliders, some different • Detector Operations begin in February, 2005! CHEP 2004
Reconstruction Requirements • Very sparse event information • Five sensors per megaton of detector CHEP 2004
Reconstruction Requirements • Reconstruction requires complicated techniques • Algorithms evolving rapidly, development distributed throughout collaboration • Software must be deployed at remote location, limited intervention possible • Desktop-to-online conversion must be easy, very robust CHEP 2004
Reconstruction Requirements Many hypotheses, hierarchical filtering Variable event flow, determined at run time CHEP 2004
Monte Carlo Requirements • Events occur at random times • No ‘heartbeat’ analogous to beam crossing • Event duration ~5 µs, rate 1700 Hz → Frequently overlapping CHEP 2004
Monte Carlo Requirements • Need to overlay events from different generators, preprocessing chains → Multiple input streams, irregular heartbeat CHEP 2004
Monte Carlo Requirements • Difficult Monte Carlo problem • Can’t track every photon from a kilometer-long PeV muon track through a km3 active volume • Run special-purpose photon tracker before Monte Carlo production, use tabulated results • Full photon tables hard to fit in memory (would use ~200 GB tables if possible) → Need to hold events in a buffer, sort for more efficient batchwise access CHEP 2004
How to Drive Execution? • Standard “push” (declarative) execution model not well suited • Individual modules need some control over the flow of the overall process • But want to avoid coupling independent code or requiring complex signalling mechanisms • “Pull” (on demand) model also problematic • Places responsibility on framework, modules to match requests, resources: developers, not users! • Enumeration of types of information that can be passed: less flexibility for developers • Either can work, but want something simple CHEP 2004
The IceTray Approach • Pushmipullyu Rube Goldberg Queue-based • Each module has one or more input, output queues (‘Inboxes’ and ‘Outboxes’) • Module receives events in an Inbox, processes them and places them in an Outbox • One Inbox is ‘active’ → drives execution • Events can be requested from the ‘passive’ Inboxes • User sets event flow topologically at configure time, arbitrary† complexity based on module logic • Order of execution determined by data flow †Required to be a Directed Acyclic Graph CHEP 2004
Inboxes and Outboxes • Context of inboxes, outboxes is created by user at configure time, not by developer CHEP 2004
Module Lifecycle • Module is a relatively flexible state machine • Default implementations are provided transparently in an inherited base class CHEP 2004
Analysis Containers • Each IceTray module isolated from larger environment, in own “container” • Interactions limited to requests for services, events, configuration, logging and errors • All services, etc. provided by Framework through a context object – switchable • Module appears to have own flow of control • Lends itself to threads, distributed processing CHEP 2004
Division of Responsibilities • Framework handles only low-level tasks • Logging, configuration, event flow, execution flow, connection of services, etc. • Does not need to understand modules, data • Modules focus exclusively on their task • No need to understand their role in the larger scheme of things • User supplies context at configure time CHEP 2004
IceTray Status • Basic implementation complete (linear functionality) • In use within IceCube for simulation, online reconstruction/data monitoring (starting this winter), offline analysis • Implementation of more advanced features in progress • Code available by request • Uses IceCube distribution, build system CHEP 2004
Old and New South Pole Stations http://glacier.lbl.gov/offline/ ICETRAY-APP/index.html CHEP 2004