Transcript APSR - Australia Telescope National Facility

APSR
Matthew Bailes
Swinburne University
Of
Technology
Baseband Pulsar Timing
History
 Caltech Berkeley Processor
 1996 10 MHz 2xDLT 7000
 Princeton Mk IV
 1997 10 MHz 2xDLT 7000
 S2TCI (~1997)
 York University/Melbourne Uni/Swinburne
 16 MHz 8xVHS
 CPSR1 (~1999)
 Caltech/Swinburne
 4xDLT 7000
Baseband History
COBRA
Coherent Baseband Recorder 2001?
150+ Processors
CPSR2 (2002)
2x64 MHz x 2bits x 2 pols
30 Xeon dual processors
Arecibo Signal Processor (ASP) 03/04?
64 MHz with polyphase filters and 4 bits
Selected Achievements
 Princeton Mk IV
 MSP timing, ~200 ns timing on 1713+0747
 PUMA2??
 CPSR1
 130 ns timing on PSR J0437-4715
 CPSR2
 Polarimetry of 27 MSPs
 Giant Pulses discovered in MSPs
 Precision timing on 7 MSPs (< 1 us)
 0437, 0613, 1600, 1713, 1744, 1909, 1937
 Several others near ~1 us
Giant Pulses
Giant Pulse
2 microseconds wide!
75 nanosecs
Timing
Stability
Profile Precision
Must be done with
Coherent dedispersion
New DFB Limitations
2048 channels/ 512 MHz
L-band
Terzan 5
190 us Smearing!
Still need coherent dedispersion
Current Generation
IF
IF
Samplers
+ FPGA
Bits
CPU 1
Polyphas
e
bits
CPU
Gb ethernet
Gb Switch
CPU
CPU
Next Generation
 “APSR”
 ATNF/Parkes/Swinburne/Recorder
 1 GigaByte/sec recorder/processor
 Much higher timing precision
 Nanosecond pulse sensitivity
 Lunar Experiments
 Giant pulses
 Pulsar searches
 Polarimetry
 Spectroscopy
 VLBI correlator?
APSR
1 GHz x 2
IF
IF
Samplers
+ FPGA
CPU x128
Bits
CPUs x16
Polyphas
e
bits
Gb ethernet x16
Gb Switch
144 ports min
Specs
Limited to 64 MB/s per host
Current Gb limit
16 Primaries
64 MB/s
2 bits x 64 MHz x 2 pols
4 bits x 32 MHz x 2 pols
8 bits x 16 MHz x 2 pols
3 GHz
L-band
Bright Pulsars
Primary Machines
3 GB RAM
Gb ethernet x 2
500 GB SATA disks
16 of them
Secondaries
Low-voltage next-gen Pentium
2 GB RAM (min)
2 x 250 GB SATA drives
Probably rack-mounted
Gb ethernet
Switch
144 ports
Gb ethernet
10 Gb uplink
Probably CISCO
Costings
Primaries: 48 K
Secondaries: 128 K
Switch: 12K
Cabling: 8K
Racks: 15K
Total: 211K
Required Modes:
 Mode 1
 Raw data x 16
 2 bits x 64 Msamples/sec x 2 pols x 16 (agg 1 GHz)
 4 bits x 32 Msamples/sec x 2 pols x 16 (agg 512 MHz)
 8 bits x 16 Msamples/sec x 2 pols x 16 (agg 256 MHz)
 Mode 2
 Polyphase FB
 2 bits x 64 MHz x 2 pols x 16 (agg 1024 MHz)
 4 bits x 32 MHz x 2 pols x 16 (agg 512 MHz)
 8 bits x 16 MHz x 2 pols x 16 (agg 256 MHz)
Required Modes:
 Mode 3
 Polyphase FB - single host
 2048 chans x 2/4/8 bits x Npol x 32 us sampling
 Up to 64 MB/s
 Mode 4
 Polyphase FB - multi-host
 16 x PPFB x nbits x nchans x Npol x Nus samples
 Mode 4
 As for 3, but incoherently dedispersed into N channels
 Mode 6
 As for 4, but coherently dedispersed
 Mode 7
 As for 3, but folded for N PSRs simultaneously
 Mode 8
 As for 6, but folded for N PSRs simultaneously
Other Wishes:
• New digital FB for next MB survey
•13 DFBs with 300 MHz BW, 2048 channels,
64 us sampling
13xIFs
GHz
Ghz
DFB
?
16?xDFB
IFs
16? lines
16 Primaries
Switch
APSR
Apple
XRAIDs
10 Gb
To Grangenet
128 CPUs
Aggregate Power
CPUs:
128 x 4GHz x 4flops/cycle = 2 Teraflops
Disk:
64 Terabytes (17 hours recording)
Pulsar surveys in real time
Issues:
Heat dissipation
144 x 150 W = 20 KW!!
Multiplexing IO to CPUs
How can I take 1 GB/sec and spread it 16 ways
without losing bits??
Upon Completion:
Throw out:
CPSR2
WBC
DFB1
MB correlators
Analogue Filters
VLBI recorders