Enabling Radio Astronomy Research Using Graphical System Design - PowerPoint PPT Presentation

1 / 40
About This Presentation
Title:

Enabling Radio Astronomy Research Using Graphical System Design

Description:

Title: Slide 1 Author: Chetan Kapoor Last modified by: Inloombp Created Date: 8/16/2006 12:00:00 AM Document presentation format: On-screen Show (4:3) – PowerPoint PPT presentation

Number of Views:90
Avg rating:3.0/5.0
Slides: 41
Provided by: Chetan65
Category:

less

Transcript and Presenter's Notes

Title: Enabling Radio Astronomy Research Using Graphical System Design


1
(No Transcript)
2
Enabling Radio Astronomy Research Using Graphical
System Design
  • Paras Loomba
  • National Instruments
  • CASPER 2011

3
Agenda
  • Needs of radio astronomy research
  • LabVIEW
  • RT-HPC?
  • Use-Case PXIe and LabVIEW for Prototyping
  • Use-Case Complex Modernization of RT-32
  • Summary

4
NeeDS of RADIO ASTRONOMY RESEARCH
5
Radio Astronomy Research Needs
  • Algorithm Development on Fast Processors
  • Large Data Collection
  • High Channel Count Reconfigurable I/O
  • Monitoring Control systems

6
Traditional Design Process
Domain Experts
Software designers
FPGA designers
ASIC/SoC designers
Custom IC designers
7
Evolution of Graphical System Design
Domain Experts
Domain Experts
Software designers
FPGA designers
System designers
8
(No Transcript)
9
Multicore Analysis Improvements and Sparse Matrix
Support
  • Enhanced multicore analysis capabilities in
    LabVIEW
  • Multithreaded linear algebra, BLAS, and FFT-based
    functions
  • Superior performance and scalability
  • Sparse linear algebra on both Windows and LabVIEW
    Real-Time
  • Both single and double precision versions of many
    functions

10
LabVIEW GPU Computing (NI Labs)
11
Upcoming GPU Analysis Offering for LabVIEW
  • Expected in 2012 Official GPU analysis library
    for LabVIEW
  • High-level API for managing GPU devices
  • Selected custom GPU kernels
  • Examples for executing custom GPU implementations
    from LabVIEW
  • Support for NVIDIA CUDA v4.0
  • Pre-wrapped CUBLAS and CUFFT functions
  • Support for development on 32-bit and 64-bit
    Windows

12
Multi-Channel FFT Example
13
What is RT-HPC?
14
Engineering Grand Challenges
Engineer the tools of scientific discovery
Advance health informatics
Reverse-engineer the brain
Provide energy from fusion
  1. Complex Computation
  2. Connection to I/O
  3. Closed-Loop Processing

Restore and improve urban infrastructure
Secure cyberspace
Provide access to clean water
Enhance virtual reality
Advance personalized learning
Develop carbon sequestration methods
Make solar energy economical
Prevent nuclear terror
Engineer better medicines
Manage the nitrogen cycle
15
Offline High Performance Computing (HPC)
Input File
Output File
Gather Data
Use Result
Sensor Data
System Feedback
16
Online Computing
Sensor Data
System Feedback
Inline Analysis Feedback
17
ESOE ELT Primary Mirror (M1) Control
42m DIAMETER10nm CORRECTION984 MIRRORS2952
ACTUATORS5904 SENSORS3k x 6k MATRIX1
MILLISECOND
M1
18
Tokamak Reactor Plasma Control
19
RT HPC System ComponentsSystem infrastructure
20
Comparing Buses
  • Bandwidth
  • The amount of data that can be transmitted in a
    given time
  • Latency
  • The time it takes from the first bit to travel
    from the transmitter to the receiver.
  • Distance
  • The maximum length the bus can span.

21
Data Bus Comparison
10000
PCI Express/ PXI Express (x4)
1000
PCI/PXI
Gigabit Ethernet
USB 2.0
100
Max Bandwidth (MB/s)
Increasing (Improving) Bandwidth
IEEE 1394a
VME/VXI
Fast Ethernet
10
GPIB (HS 488)
USB 1.1
GPIB (488.1)
1
1
10
100
1000
10000
0.1
Approximate Latency (µs)
Decreasing (Improving) Latency
22
NI PXIe-1082 Chassis Backplane
7 GB/s Total System Data Bandwidth
23
RT HPC System ComponentsInterconnect bus
24
Specification
  • PXI MultiComputing Specification
  • In development within the PXI Systems Alliance
    (PXISA)
  • Leverages the PCI(e) NTBs to connect intelligent
    end-points
  • Hardware and Software specifications to define a
    vendor interoperable solution
  • Hardware
  • Base NTB requirements
  • Cabling
  • Software
  • Standard API

25
Natural Extension of PXI Express
  • PXImc extends PXI Express by
  • Increasing Processing Capabilities
  • Enabling Distributed Topologies
  • Reducing Bandwidth Bottlenecks
  • Compatible with PXI Express
  • No changes to existing chassis, controllers, or
    modules
  • PXImc enables new PXI applications
  • Hybrid High Performance TM Systems (ex. HIL,
    Real-Time Test)
  • Distributed Processing via Multicore CPUs (ex.
    SIGNIT)
  • In-system Co-processing via x86 based CPUs

26
Hybrid High Performance TM Systems
Master PXI System
Cabled PCI Express Links
PCIe Switch
Secondary PXI System
Secondary PXI System
High Performance Boxed Instrument
27
Distributed Processing via Multicore CPUs (x86)
Cabled PCI Express Links
28
In-system Co-processing via x86 based CPUs
PXI Chassis with Multiple Processors
Master Controller
I/O Module
I/O Module
I/O Module
I/O Module
Processor Module
I/O Module
Processor Module
Communication to from System Controller
Peer-to-Peer Streaming between Processor Module
I/O Module
29
RT HPC System ComponentsHeterogeneous computing
30
Different Processing Technologies
Technology Advantages
Multi-Core CPUs Floating Point Operations Diversity of Tasks Some Parallelism
FPGAs Direct Connection to I/O for In-Line Processing High Parallelism High Throughput (fixed-point operations)
GP-GPUs Potentially High Throughput and Parallelism (for specific tasks)
(Images courtesy of Intel, Xilinx, NVIDIA)
31
Heterogeneous Computing Elements
Sensor Data
System Feedback
Inline Analysis Feedback
Multi-Core CPUs
GP-GPUs
FPGAs
(Images courtesy of Intel, Xilinx, NVIDIA)
32
Using PXI and LabVIEW
33
Example RTHPC System with PXI Platform
Hardware Components
Software Components
  • LabVIEW
  • High-Performance Analysis Library (HPAL)
  • Sparse Linear Algebra Library
  • LabVIEW FPGA
  • CUDA/CUBLAS Libraries
  • Modular Instruments (I/O)
  • PXI Embedded Controller
  • FlexRIO FPGAs
  • Peer-to-Peer Streaming
  • Rack-Mount Controllers
  • PXImc
  • NVIDIA Tesla GP-GPU
  • Cable PCI-Express

NVIDIA Tesla S1070 (4 GP-GPU)
NI 8354 (Multicore RAID 0)
(Image courtesy of NVIDIA)
34
USE Cases GBT ,NRAO
35
PXI and LabVIEW FPGAPrototyping Algorithms
36
PXI and LabVIEW FPGAPrototyping Algorithms
  • Used LabVIEW to Program Onboard FPGA
  • Developed Algorithms for Side band Seperation and
    Polarization
  • PXIe bus to stream Data to 12TB RAID Disk

37
USE Cases mc OF rt - 32
38
Monitoring and Control Smart SCADA
39
Monitoring and Control Smart SCADA
  • Power Subsystem Automation
  • Antennae control system (ACS)
  • Radio measurements Storage Subsystem
  • Telescope Physical Parameters
  • Auxiliary Subsystem for Optical Guidance by Stars

40
Summary
  • Multi Domain - LabVIEW
  • Evolution of PXI to PXImc
  • SMART Monitoring Control on
  • NI RT controllers
Write a Comment
User Comments (0)
About PowerShow.com