IT/EE Palaver - 20.06.06FAIR DAQ - J.Adamczewski, S.Linev1.

Slides:



Advertisements
Similar presentations
Operating System.
Advertisements

Middleware Support for RDMA-based Data Transfer in Cloud Computing Yufei Ren, Tan Li, Dantong Yu, Shudong Jin, Thomas Robertazzi Department of Electrical.
The Development of Mellanox - NVIDIA GPUDirect over InfiniBand A New Model for GPU to GPU Communications Gilad Shainer.
AMLAPI: Active Messages over Low-level Application Programming Interface Simon Yau, Tyson Condie,
Setting up of condor scheduler on computing cluster Raman Sehgal NPD-BARC.
CHEP 2012 – New York City 1.  LHC Delivers bunch crossing at 40MHz  LHCb reduces the rate with a two level trigger system: ◦ First Level (L0) – Hardware.
Remigius K Mommsen Fermilab A New Event Builder for CMS Run II A New Event Builder for CMS Run II on behalf of the CMS DAQ group.
VIA and Its Extension To TCP/IP Network Yingping Lu Based on Paper “Queue Pair IP, …” by Philip Buonadonna.
Rheeve: A Plug-n-Play Peer- to-Peer Computing Platform Wang-kee Poon and Jiannong Cao Department of Computing, The Hong Kong Polytechnic University ICDCSW.
ACAT 2002, Moscow June 24-28thJ. Hernández. DESY-Zeuthen1 Offline Mass Data Processing using Online Computing Resources at HERA-B José Hernández DESY-Zeuthen.
Timm M. Steinbeck - Kirchhoff Institute of Physics - University Heidelberg 1 Timm M. Steinbeck HLT Data Transport Framework.
Embedded Transport Acceleration Intel Xeon Processor as a Packet Processing Engine Abhishek Mitra Professor: Dr. Bhuyan.
An overview of Infiniband Reykjavik, June 24th 2008 R E Y K J A V I K U N I V E R S I T Y Dept. Computer Science Center for Analysis and Design of Intelligent.
L. Granado Cardoso, F. Varela, N. Neufeld, C. Gaspar, C. Haen, CERN, Geneva, Switzerland D. Galli, INFN, Bologna, Italy ICALEPCS, October 2011.
Architectural Software Support for Processing Clusters Johannes Gutleber, Luciano Orsini European Organization for Nuclear Research Div. EP/CMD, The CMS.
German National Research Center for Information Technology Research Institute for Computer Architecture and Software Technology German National Research.
CERN - IT Department CH-1211 Genève 23 Switzerland t The High Performance Archiver for the LHC Experiments Manuel Gonzalez Berges CERN, Geneva.
Data Acquisition Backbone Core DABC J. Adamczewski, H.G. Essel, N. Kurz, S. Linev GSI, Darmstadt The new Facility for Antiproton and Ion Research at GSI.
D ata A cquisition B ackbone C ore J.Adamczewski, H.G.Essel, N.Kurz, S.Linev 1 Work supported by EU.
Test Of Distributed Data Quality Monitoring Of CMS Tracker Dataset H->ZZ->2e2mu with PileUp - 10,000 events ( ~ 50,000 hits for events) The monitoring.
High Performance User-Level Sockets over Gigabit Ethernet Pavan Balaji Ohio State University Piyush Shivam Ohio State University.
Artdaq Introduction artdaq is a toolkit for creating the event building and filtering portions of a DAQ. A set of ready-to-use components along with hooks.
A TCP/IP transport layer for the DAQ of the CMS Experiment Miklos Kozlovszky for the CMS TriDAS collaboration CERN European Organization for Nuclear Research.
Boosting Event Building Performance Using Infiniband FDR for CMS Upgrade Andrew Forrest – CERN (PH/CMD) Technology and Instrumentation in Particle Physics.
1-1 Embedded Network Interface (ENI) API Concepts Shared RAM vs. FIFO modes ENI API’s.
The MPC Parallel Computer Hardware, Low-level Protocols and Performances University P. & M. Curie (PARIS) LIP6 laboratory Olivier Glück.
Module 10: Monitoring ISA Server Overview Monitoring Overview Configuring Alerts Configuring Session Monitoring Configuring Logging Configuring.
C.Combaret, L.Mirabito Lab & beamtest DAQ with XDAQ tools.
ATCA based LLRF system design review DESY Control servers for ATCA based LLRF system Piotr Pucyk - DESY, Warsaw University of Technology Jaroslaw.
Multiple Processor Systems Chapter Multiprocessors 8.2 Multicomputers 8.3 Distributed systems.
Swapping to Remote Memory over InfiniBand: An Approach using a High Performance Network Block Device Shuang LiangRanjit NoronhaDhabaleswar K. Panda IEEE.
Application Block Diagram III. SOFTWARE PLATFORM Figure above shows a network protocol stack for a computer that connects to an Ethernet network and.
D ata A cquisition B ackbone C ore DABCDABC , Huelva J.Adamczewski, H.G.Essel, N.Kurz, S.Linev 1 Work.
Frank Lemke DPG Frühjahrstagung 2010 Time synchronization and measurements of a hierarchical DAQ network DPG Conference Bonn 2010 Session: HK 70.3 University.
Upgrade of the CMS Event Builder Andrea Petrucci - CERN (PH/CMD) on behalf of the CMS DAQ group 19 th International Conference on Computing in High Energy.
DABCDABC J. Adamczewski-Musch, H.G. Essel, S. Linev Software development for CBM DAQ J. Adamczewski-Musch, H.G. Essel, S.
7. CBM collaboration meetingXDAQ evaluation - J.Adamczewski1.
Management of the LHCb DAQ Network Guoming Liu * †, Niko Neufeld * * CERN, Switzerland † University of Ferrara, Italy.
Scalable Systems Lab / The University of New Mexico© Summer 2000 by Adrian Riedo- Slide 1 - by Adrian Riedo - Summer 2000 High Performance Computing using.
Data Acquisition Backbone Core J. Adamczewski-Musch, N. Kurz, S. Linev GSI, Experiment Electronics, Data processing group.
An Architecture and Prototype Implementation for TCP/IP Hardware Support Mirko Benz Dresden University of Technology, Germany TERENA 2001.
Developments and applications of DAQ framework DABC v2 Jörn Adamczewski-Musch, Nikolaus Kurz, Sergey Linev GSI / Experiment Electronic.
DABCDABC D ata A cquisition B ackbone C ore J.Adamczewski, H.G.Essel, N.Kurz, S.Linev 1 Work supported by EU RP6 project.
System Components ● There are three main protected modules of the System  The Hardware Abstraction Layer ● A virtual machine to configure all devices.
1 MSRBot Web Crawler Dennis Fetterly Microsoft Research Silicon Valley Lab © Microsoft Corporation.
CSI 3125, Preliminaries, page 1 SERVLET. CSI 3125, Preliminaries, page 2 SERVLET A servlet is a server-side software program, written in Java code, that.
CMS Luigi Zangrando, Cern, 16/4/ Run Control Prototype Status M. Gulmini, M. Gaetano, N. Toniolo, S. Ventura, L. Zangrando INFN – Laboratori Nazionali.
DABCDABC D ata A cquisition B ackbone C ore J.Adamczewski, H.G.Essel, N.Kurz, S.Linev 1 Work supported.
DABC Data Acquisition Backbone Core NUSTAR, Legnaro : DABC - J.Adamczewski, H.G.Essel, N.Kurz, S.Linev 1 Data Acquisition Backbone Core J.Adamczewski,
1 Farm Issues L1&HLT Implementation Review Niko Neufeld, CERN-EP Tuesday, April 29 th.
Management of the LHCb DAQ Network Guoming Liu *†, Niko Neufeld * * CERN, Switzerland † University of Ferrara, Italy.
CMS Luigi Zangrando, Cern, 16/4/ Run Control Prototype Status M. Gulmini, M. Gaetano, N. Toniolo, S. Ventura, L. Zangrando INFN – Laboratori Nazionali.
COMPASS DAQ Upgrade I.Konorov, A.Mann, S.Paul TU Munich M.Finger, V.Jary, T.Liska Technical University Prague April PANDA DAQ/FEE WS Игорь.
Remigius K Mommsen Fermilab CMS Run 2 Event Building.
Online Software November 10, 2009 Infrastructure Overview Luciano Orsini, Roland Moser Invited Talk at SuperB ETD-Online Status Review.
The Evaluation Tool for the LHCb Event Builder Network Upgrade Guoming Liu, Niko Neufeld CERN, Switzerland 18 th Real-Time Conference June 13, 2012.
Scalable Readout System Data Acquisition using LabVIEW Riccardo de Asmundis INFN Napoli [Certified LabVIEW Developer]
DHH at DESY Test Beam 2016 Igor Konorov TUM Physics Department E18 19-th DEPFET workshop May Kloster Seeon Overview: DHH system overview DHE/DHC.
An open source user space fast path TCP/IP stack and more…
Advisor: Hung Shi-Hao Presenter: Chen Yu-Jen
Towards a High Performance Extensible Grid Architecture Klaus Krauter Muthucumaru Maheswaran {krauter,
HTCC coffee march /03/2017 Sébastien VALAT – CERN.
Enhancements for Voltaire’s InfiniBand simulator
J. Gutleber, L. Orsini, 2005 March 15
RT2003, Montreal Niko Neufeld, CERN-EP & Univ. de Lausanne
Chapter 3: Windows7 Part 4.
MPJ: A Java-based Parallel Computing System
Network Processors for a 1 MHz Trigger-DAQ System
Chapter 13: I/O Systems I/O Hardware Application I/O Interface
STATEL an easy way to transfer data
Presentation transcript:

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev1

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev2 Outline Requirements of DAQ for CBM The InfiniBand Cluster, uDAPL XDAQ tests on the IB-Cluster Summary and further evaluation

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev3 CBM data acquisition data dispatcher FEE deliver time stamped data CNet collect data into buffers Detector collect ~50000 FEE chips event dispatcher 1000x1000 switching BNet sort time stamped data ~1000 links a 1 GB/sec HNet high level selection to high level computing and archiving ~1 GB/sec Output processing PNet process events level 1&2 selection subfarm ~100 subfarms ~100 nodes per subfarm ~10 dispatchers → subfarm ~1000 collectors ~1000 active buffers TNet time distribution

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev4 CBM DAQ features Triggerless data transport until filter farm Event building on full data rate ~1TB/s B-net: ~1000 nodes, high-speed interconnections Linux may run on all DAQ nodes (even FPGAs) Test cluster with InfiniBand: small „demonstrator“ set-up within next year

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev5 InfiniBand test cluster 4 nodes: –Double Opteron 2.2 GHz, 2GB RAM –Mellanox MHES18-XT host adapter –2x Gigabit Ethernet host adapters –SuSE Linux 9.3, x64bit version Mellanox MTS X InfiniBand switch PCI Express x8

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev6 Testcluster networks InfiniBand switch: MTS2400 InfiniBand HCA: MHES18-XT Gigabit Ethernet switch Ethernet cabling InfiniBand cabling

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev7 InfiniBand software Mellanox IB Gold –IPoIB: IP over InfiniBand driver –uDAPL: User Direct Access Programming Layer –MPI: Message Passing Interface –others...

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev8 Direct Access Transport Developed by DAT collaborative Objectives: –Transport and platform (OS) independence –Define user- (uDAPL) and kernel-level (kDAPL) APIs Features: –Zero-copy data transfer –RDMA - Remote Direct Memory Access

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev9 uDAPL test application C++ wrapper for C-based uDAPL library Same program runs on all nodes All-to-all connectivity Time synchronization between nodes ~2 µs One master (deliver commands) and many slaves Several kinds of traffic schedule can be generated and executed Statistics like transfer rate, lost packets, schedule accuracy can be measured

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev10 All-to-all communication test Each node able to send data to any other node Message- and RDMA-based transfer are supported Master generate schedule and distribute it to all slaves At predefined time schedule execution is started Several schedules were tested: –One-to-all schedule –All-to-one schedule –All-to-all round-robin –All-to-all fix target masternode02node01node03

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev11 Scheduled test results Slight difference between message- and RDMA-based transfer Reasonable transfer rates starting from 16 KB buffers

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev12 Chaotic (schedule-less) transfer tests Packets send without schedule Next packet send when previous is completed Performance depends on receiving queue size For small packets performance 2 times better than scheduled approach Low CPU usage

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev13 Data Acquisition framework requirements Configuration of multiple nodes (connections and algorithms, database?) Controls and monitoring Data transport (format, protocol, hardware drivers?) Message logging Error handling and failure recovery (ARMOR?) Modular architecture, extensability! (sub-detector tests) User interface?

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev14 Standard DAQ framework for LHC CMS experiment (Orsini, Gutleber) C++ libraries on Linux, modular packages Each node: context, xdaq process with embedded xdaq applications Configuration: XML Data transport: I2O protocol (Intelligent IO) Communication: http, cgi; SOAP messages The CMS XDAQ framework

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev15 XDAQ features continued... State machines (sync/async FSM) Message logger, error handler Monitoring tool Hardware access library (HAL) Front End Driver (FED kit, for CMS!) Job Control (task handler for node control) others: exceptions threads infospace data (de)serializers,... The CMS XDAQ framework

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev16 XDAQ on the CBM InfiniBand cluster Installation problems: XDAQ distribution not 64 bit save! Code adjustments required kernel modules for CERN linux excluded! drivers for CMS hardware (FED) excluded coretools done, worksuite partially,…  Basic evaluation possible!

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev17 XDAQ tests on CBM InfiniBand cluster Tested features on IB cluster: hyperdaq webserver (main test controls ui) xml configuration setup control variable access and export (infospace) SOAP messaging (test controller application) state machines (sync./async. with web/SOAP ui) monitoring application (very raw ui!) multithreading (workloop framework) exceptions, data wrappers, message logging, toolbox,… i2o data transport (tcp roundtrip example)

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev18 XDAQ: web server hyperdaq web interface

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev19 XDAQ executive with applications Common XML configuration file: Each node knows all applications on all nodes! Unique addressing by context (url) and application id

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev20 XDAQ: SOAP messaging xrelay: send SOAP control messages example commands: Configure, Enable, Halt, Reset,.. any new commands may be defined web interface not suited as real control system UI! XDAQ applications may exchange SOAP messages with other UI (Labview?)

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev21 XDAQ: messaging and peer transport peer transport i2o tcp fifo http SOAP messages transport layer DAPL new development

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev22 XDAQ: I2O messaging and peer transport

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev23 Peer Transport DAPL (ptdapl) GSI development 03/ /2006 C++ wrapper class for uDAPL C library (class TBasic) peer transport tcp as „template“ (starting point) uDAPL buffers managed within XDAQ memory pool (class TEndpointBuffer) avoids memcopy and new buffer allocation for each send package: l ookup if posted memory reference is known as send buffer user code can write directly into uDAPL send buffer multiple threads for sending, releasing, and receiving buffer

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev24 Peer Transport DAPL structure uDAPL peer transport senderreceiver transmitter messenger receiverloop buffer pool message (I2O) channel

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev25 I2O roundtrip on InfiniBand cluster Standard XDAQ roundtrip benchmark: sender node posts I2O frame to receiver node receiver callback sends frame back to sender sender callback sends frame back again, etc… perfmeter class records bandwidth, latency,… web display of results on sender node (java!) Code independent of transport implementation! XML configuration file does set up (pttcp eth, pttcp IB, ptDAPL, ptFIFO,...)

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev26 XDAQ executive MyRoundTrip HyperDAQ I2O roundtrip setup sender receiver XDAQ executive MyRoundTrip token() EnableAction() default() I2O callback web application state machine Peer Transport

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev27 I2O sender/receiver benchmark New applications MyDataSource, MyDataDrain: sender node posts I2O frames to receiver node in workloop thread receiver callback gets frame and checks it multiple senders/receivers possible (n x m) perfmeter class measures on all nodes web display of results on all nodes (java!) Code independent of transport implementation! XML configuration file does set up (pttcp eth, pttcp IB, ptDAPL, ptFIFO,...)

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev28 HyperDAQ I2O sender/receiver setup senderreceiver XDAQ executive MyDataSource Benchmark() EnableAction() default() workloop (thread) web application state machine XDAQ executive MyDataDrain token() I2O callback Peer Transport default()

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev29 Transport bandwidth B and latency  P: package size  : transfer time („latency“) Observation:  linear with P:  B P P P big: network limit P small: framework latency !

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev30 ptDAPL benchmark results TestNbufs,rcv  0 (µs) C (µs/kByte) 1/  0 (kHz) Bmax (MByte/s) MyRoundTrip MySourceDrain MySourceDrain MySourceDrain MySourceDrain

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev31 Benchmark summary 955 MB/s plain uDAPL 234 MB/s XDAQ, I2O, TCP over IB 905 MB/s XDAQ, I2O, ptDAPL

IT/EE Palaver FAIR DAQ - J.Adamczewski, S.Linev32 Summary and Outlook CBM DAQ requires fast builder network: InfiniBand test cluster IB hardware and software installed, tested,... uDAPL: developed C++ wrapper and test applications many tests with different traffic patterns were performed more tests on bigger InfiniBand clusters are required XDAQ as software framework: tested features: SOAP, web server, data transport,... I2O implementation for uDAPL done! cluster configuration? control system? scalability? hardware access?