1. The High Time Resolution Universe Survey Backend
Ensure you thank relevant students/postdocs etc.
Bailes, van Straten, Jameson, McMahon, Werthimer, Keith.

2. Motivation
Pulsars
Surveys
The multiBOB
Processing
Results
MeerKAT

3. Magnetic Field - Period Diagram
Crab pulsar
“Noisy”
Pulsars go
this way
“Noisy”
“Smooth”
Pulsar
recycling

4. Parkes Multibeam Surveys
Discovered 850+ pulsars
Double pulsar (0737)
NS+NS binaries (1756)
Magnetar-like pulsars
NS+WD pulsars (1141)
Superb timing pulsars (1909)
Had poor frequency resolution

5. Galactic Distribution
Need to hit the plane

6. DM contours & known MSPs
Free Electron Density

7. Pulse Dispersion

8. Pulsar Dispersion Smearing
DM=100, B=3 MHz, 1350 MHz, t=1.1ms

9. PSR J
40 s wide!!

11. Single pulses: the past 2 years
30 Jy one-off 5ms burst
DM = 375
Not seen again in 90 hours observing.
What is it?
Extreme RRAT?
Supernova event?
Coalesced NS-NS?
Black hole evaporation?
Cleverly dispersed RFI?
Aliens?
tdm f-2
tscat f-4
THE MAIN FOCUS OF OUR SEARCH
WHAT MAKES IT SEEM LEGIT
DISTANCE ESTIMATE
WHAT IS IT?
Lorimer et al. 2007

15. MultiBOB iBOB Xeon Disk CPU cluster casper psrdada linux
iBOB + 10 Gb ethernet + server TB disk
iBOBs integrate FFTs on each pol for 64 us
We select best 8 bits on each pol
Talk over 10 Gb
iBOB
10Gb
Xeon
Disk
CPU cluster
casper
psrdada
linux
Swinproc/sigproc

16. Introducing: BPSR MultiBOB 13 beam spectrometer
1024 channels, 64 us sampling, 3.3K each!
High Time Resolution Universe Survey
Request:
~200 days of Parkes time over 4 years
Discover 100s more pulsars
Fastest pulsars
Gb link to Green Machine

20. Procedure Form average bandpasses from first 10s
2-bit digitize thereafter
Write to file + sigproc (Lorimer) header

22. A tale of two surveys PKS Multibeam Tsamp = 250 us ChBW = 3 MHz
Nbit = 1
5 TB
High Time Resolution Universe
Tsamp = 64 us
ChBW = 0.4 MHz
Nbit = 2
~1000 TB (150K in tapes)

23. A tale of two surveys PKS Multibeam Tsamp = 250 us ChBW = 3 MHz
Nbit = 1
5 TB
High Time Resolution Universe
Tsamp = 64 us
ChBW = 0.4 MHz
Nbit = 2
~500 TB

25. The Green Machine Supercomputing @ Swinburne installed May/June 2007
185 Dell PowerEdge1950 nodes
2 quad-core processors
(Clovertown: Intel Xeon 64-bit 2.33 GHz)
16GB RAM
1TB disk -> 300 TB total
1640 cores/14 Tflops
20 Gb infiniband (Q1 2009)
83 kW .vs. 130 kW cooling
Machines: ~1.2M
Fuel: ~100K/yr

26. DM=297
P=2.149ms

29. HITRUN processing Requires 1040 Clovertown cores for real time
Debird(transpose): 3 mins + 10 mins
Dedisperse: 80 mins
FFT: 15 mins
Fold: 60 mins
Optimize: 5 mins
Total: ~3 hours/ 4cores

30. ? Supercomputing 1 Teraflops for 1K 128 GB/s RAM BW 1M = 128 TB/s
1 Petaflop
300 kW!!! ?

31. HITRUN processing GPU processing 1 Tesla S1070 has 4xGPUs
Debird (transpose): 3 mins + 10 mins, 1.5 sec + 17 sec (FFT)
Dedisp: 80 mins, 5 mins
FFT: 15 mins, 30 sec
Fold: 60 mins, ???
Optimize: 5 mins, ???
Total: ~3 hours/ 4cores, 10 min
1 Tesla S1070 has 4xGPUs
Equivalent to 36 servers or 288 cores (10K)
Real time search with 4 S1070s? (40K)
Supermicro make a Nehalem + 2 x Tesla
10 of these ~100K

32. Storage vs Real time processing
1,000,000,000 MB / (53 MB/s) = 220 days
1,000,000,000 MB = 150K / copy
GPU farm ~ 100 K
Observers
200K
Parkes Telescope
2,000 K

33. 1623-4949B ~4 Second pulsar DM ~1000 300 sigma! B - 1014 G!
Profile changes
Timing noise

34. 1623 humps

35. 1623 position

36. Operations MultiBOB comms sensitive. MultiBOBs hate LO changes.
No failures in ~year (x13).
Combination of software & hardware good
Didn’t know algorithm without trying
Wish-list
“Saturation”

38. MeerKAT as a pulsar finder?
In total power
Like a 36 m dish
FoV like a 12 m ~ 30 x PKS
G sqrt(Bt)
PKS MB = 1
MeerKAT = 0.5
Ultra-compact coherent machine?
8 x 8 2D (32x32) 2D FFTs (300m x 300m)
#Telescope beams? ~1024 beams
MeerKAT = 2.1 (gain/T) x 1.7(sqrt t) = ~3xPKS

39. Data Rates? 1024 channels 2 bits 1024 = 32 x 32 beams 8 kHz sampling
4 GB/s cf Raw rate of 81 x 1.2 GB/s/telescope
300 MHz BW

40. Real time GPU/CPU Petaflop processor
@8 kHz, 2x faster real time/beam
Next-gen GPU (2Tflops), 4x faster?
Need 256 GPUs
128 Supermicros + 2xGPU
Only 1.3 M