220 likes | 316 Vues
Explore the challenges and solutions of the embedded distributed system of Antares neutrino telescope, balancing hardware constraints and software efficiency. Learn about component integration, network topologies, power consumption, and deployment constraints affecting the detector control and data acquisition. Discover the technologies and programming languages used in this complex system.
E N D
Addressing the Software and Hardware Trade-Offs of an Embedded Distributed System The Case of the ANTARES Data Acquisition and Detector Control S. Anvar, CEA Saclay Dapnia On behalf of the Antares collaboration
A Neutrino Telescope http://antares.in2p3.fr RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
The Antares Collaboration • Institutes • Particle physics • Astronomy • Sea science RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Constraints • Environmental constraints • Baseline data rate of ~100 kHz / PMT • Data rate fluctuations ~2MHz / PMT due to bioluminescence • Strong uncertainties on environment parameters • Reduced space and thermal conductivity • … • Reliability (MTBF) • Maximize simplicity • Minimize consumption • Minimize number of components • … • Deployment constraints • High cost of submarine cables • Limits on string cable rigidity or sea operations • … RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Slow control Data Clock Energy Design principle Offshore Onshore Electro-mechanicalcable • 12 strings • 25 detecting nodes per string (every 12.5m) • 3 optical modules (PMTs) per acquisition node • 2 "ARS" digitizing chips per optical module Processingnodes Acquisitionnodes • Real time processing: • Offshore: readout 1800 ARS chips spread out over 30 000 000 m3 @ 2500m underwater • Onshore: perform real time trigger computations on incoming data. String control Detectors Junctionbox Electro-opticalcable (50 km) RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Ethernet 100 Serial links (RS485, RS232) Each offshore acquisition node: one DAQ/SC Board • Achieved power consumption of ~5W including optical Ethernet tranceiver laser DAQ/SC Board ARS readout/SCLocal trigger106 gatesXilinx FPGA SDRAMMemory 64 MB Real-timeembedded CPUPPC MPC860VxWorks RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
The ARS chip and the DAQ/SC board DAQ/SC Board ARS chip RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
100 Mb/s M Sectors 1 Gb/s M To Shore (DWDM) SCM SCM Network Topology (1) Cable rigidity Cable diameter Connection complexity Each LCM has its own I/O data flow(Slow Control & Physics Data) RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Network Topology (2) RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
The offshore switch board Offshore switch Board Allayer chip RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
The electronic crate RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
N N N N N Network Topology (3) • Solution • barrel shifting • synchronized by global clock N N N N N ~15 Mb/s SECTOR SWITCH (in: 5x100, out: 1x1000) SECTOR SWITCH (in: 5x100, out: 1x1000) Used as data concentratormany input ports one output port risks of congestion intelligent flow control in node ON-SHORE SWITCH: (in: 60x1000, out: 20x1000) Instant ~6 Gb/s PC PC PC PC PC PC PC RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Massively Distributed and Heterogeneous System • Final System: A Network of ~320 Nodes • ~300 Offshore Nodes (MPC860 + VxWorks) • ~20 Onshore Nodes (PC + Linux) • A number of distributed applications running in parallel and communicating • DAQ • L2 Trigger • Slow Control / Acoustic positioning acquisition • L1 trigger ? • Run control RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Software Technologies • Programming Languages • C/C++ for offshore software components • C++ / Java for onshore processing • Java / C++ for onshore display • Distribution • Multiple messaging and data servers (ControlHost) TCP/IP based • VxWorks Zero-copy TCP/IP (~25Mb/s ~38 Mb/s per node) • Offshore network traffic control (barrel shifting) • Data and configuration storage • Oracle database • Root files • Design & Development • Object Paradigm • UML (Unified Modeling Language) + Extensions • Model-driven engineering RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Interrupt – shared memory communication Data Memory (SDRAM) ARS readout (FPGA) write intrHandler (RTOS) frameSendertask (RTOS) Semaphore(RTOS) read and write Actual frame formatting and sending through TCP/IP socket RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
asynchronous method call :ARS_Readout :FrameSender reference to design pattern acq(in frameID:int) Interrupt - SharedMem 6 1 <<RTOS>> FS_Node <<fpga>> AR_Node FrameSender ARS_Read readout framer +acq(in frameID:int) +fillBuffers(id:int) 6 1 frameSender <<reside>> <<reside>> ARS_Readout FrameSender Interrupt – shared memory communicationseen as asynchronous method call RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Implemented in VHDL RealTime::Thread #run() PrmRefType IntrSharedMemSkeleton #run() #installInterrupt() +accept(in pr:PrmRefType) ARS_Read FrameSender FrameSenderSkeleton +fillBuffers(id:int) +acq(in x:Frame) {bind(Frame)} +accept(in x:Frame) IntrSharedMemProxy +doIntr(in x:PrmRefType) Applying the design pattern: model transformation 6 reusablesoftware 1 1 frameSender RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Widespread belief Performance goes against abstraction & flexibility Wrong if the issue is explicitly addressed ActuallyAbstraction modularity specific optimization perf. RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
DAQ/SC Board ARS readout/SCLocal trigger106 gatesXilinx FPGA SDRAMMemory 64 MB DAQ/SC Board ARS readout/SCLocal trigger Real-timeembedded CPU System On Chip Xilinx Virtex II Pro?Xilinx Virtex IV Real-timeembedded CPUPPC MPC860VxWorks SDRAMMemory 64 MB Ethernet 100 Serial links (RS485, RS232) Ethernet 100 Serial links (RS485, RS232) Evolutions (1) • Lower consumption • Reduced number of components increased reliability RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Evolutions (2) • Present scheme: « all data to shore » • Unexpected variable baseline rates: ~100 400 kHz instead of ~70 RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Present scheme: « all data to shore » Unexpected variable baseline rates: ~100 400 kHz instead of ~100 Evolutions (2) From 5 to 30 cm/s (!) • Software L1 trigger • Reduced L1 partial data sent from offshore to central L1 workstation • Central L1 computes trigger with global vision of detector • Trigger decision broadcasted through software • Expected gain in data rate: x1000 RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control
Slow control Data Clock Energy Conclusions • HEP DAQ and trigger techniques • High data rates • Very weak signal • Highly distributed system • Soft real time constraints • … • Challenges unusual for HEP • No known beam (calibration) • Very embedded system • Very hostile environment • Uncontrolled factors (biolum) • … • Both Robust and Flexible Design • Allows adaptation • Allows for remote evolution • … RT2005 – S. Anvar – CEA Dapnia – Adressing the HW and SW trade-offs of an embedded distributed system: the case of the Antares DAQ and detector control