EXPReS FABRIC WP 2.2 Correlator Engine Meeting 25-09-2006 Poznan Poland JIVE, Ruud Oerlemans 25-09-2006 EXPReS FABRIC meeting at Poznan, Poland 1 WP2.2 Correlator Engine Develop a Correlator Engine that can run on standard workstations, deployable on clusters and grid nodes 1. 2. 3. 4. 5. 6. 7. Correlator algorithm design (m5) Correlator computational core, single node (m14) Scaled up version for clusters (m23) Distributed version, middle ware (m33) Interactive visualization (m4) Output definition (m15) Output merge (m24) 25-09-2006 EXPReS FABRIC meeting at Poznan, Poland 2 Current broadband Software Correlator Station 1 Station 2 Station N EVN Mk4 equivalents Raw data BW=16 MHz, Mk4 format on Mk5 disk From Mk5 to linux disk Raw data 16 MHz, Mk4 format on linux disk DIM,TRM, CRM Channel extraction Extracted data SU Pre-calculated,Delay tables DCM,DMM, FR Delay corrections Delay corrected data Correlator Chip Correlation. SFXC Data Product 25-09-2006 EXPReS FABRIC meeting at Poznan, Poland 3 High level design Distributed Correlation Principle Investigator SCHED schedule EOP VEX Telescope operator WFM CALC Field System Process VEX Mark5 System VEX Central operator CCF Delay Field System Mark5 System Grid Node Field System Mark5 System 25-09-2006 Grid Node Grid Node EXPReS FABRIC meeting at Poznan, Poland JIVE archive 4 Grid considerations/aspects Why use grid processing power? It is available, no hardware investment required It will be upgraded regularly Degree of distribution is trade-off between Processing power at the grid nodes Data transport capacity to the grid nodes Data logistics and coordination More complicated when more distributed Processing at telescope and grid nodes Station related processing at telescope site and correlation elsewhere All processing at grid nodes 25-09-2006 EXPReS FABRIC meeting at Poznan, Poland 5 Data distribution over grid sites (1) Baseline slicing Pros •Small nodes •Simple implementation at node Cons •Multiplication of large data rates, especially when number of baselines is large •Data logistics complex •Scalability complex 25-09-2006 EXPReS FABRIC meeting at Poznan, Poland 6 Data distribution over grid sites (2) Pros •Simple data logistics •Central processing •Live processing easy •Slicing at the grid site •Dealing with only one site. Cons •Powerful central processing site required 1 All data to one site Time slicing Pros •Smaller nodes •Live processing possible •Data slicing at nodes Cons •Multiplication of large data rates •Simultaneous availability of sites when processing live All data to different sites Pros •Smaller nodes •Live processing per channel •Simple implementation •Easy scalable Cons •Channel extraction at telescope increases data rate Pros •Smaller nodes •Smaller data rates •Simple implementation •Easy scalable •No data mulitplication Cons •Complex data logistics after correlation •Live correlation complex 3 25-09-2006 2 Channel slicing EXPReS FABRIC meeting at Poznan, Poland 4 7 Correlator architecture for file input Time slice 1 SA SB SC Time slice 2 SA SB SC Time slice 3 SA SB SC Core1 CP1 •Processes data from one channel •Easy scalable, because one application has all the functionality •Can exploit multiple processors using MPI •Code reuse through OO and C++ Core2 CP2 Core1 SD SD SD Core3 CP3 CP This software architecture can work for data distributions 1,2 and 3 Offline processing 25-09-2006 EXPReS FABRIC meeting at Poznan, Poland 8 Correlator architecture for data streams 1.1 CP 1.1 1.2 1.2 SA SB SC SD 1.3 2.1 Core1 1.3 2.1 Buffer 1 Time 2.2 2.2 2.3 3.1 Core2 Buffer 2 Core4 3.2 2.3 3.1 3.2 3.3 Core3 4.1 Buffer 3 3.3 4.1 4.2 4.2 4.3 4.3 Memory buffer with short time slices 25-09-2006 Real time processing EXPReS FABRIC meeting at Poznan, Poland File on disk 9 Other issues Swinburne University, Adam Deller Last summer exchange of expertise on their software correlator New EXPReS employee: Yurii Pidopryhora, Astronomy background Data analysis and testing New SCARIe employee: Nico Kruithof Computer science background Scari, NWO funded project aimed at sw correlator on Dutch Grid 25-09-2006 EXPReS FABRIC meeting at Poznan, Poland 10 WP 2.2.? Status Work Package 1. Correlator algorithm design 2. Correlator computational core 3. Scaled up version for clusters 4. Distributed version 5. Interactive visualization 6. Output definition 7. Output merge 25-09-2006 M 5 14 23 33 4 15 24 Status almost finished active active pending pending designing designing EXPReS FABRIC meeting at Poznan, Poland 11
© Copyright 2025 Paperzz