Biomedical Analysis and Simulation Supercomputer

1
Biomedical Analysis and Simulation Supercomputer

• Systems MS RA work
• (Graphics and Image folks can nap)
• (Ph.D. intending folks can nap)

2
Biomedical Analysis and Simulation Supercomputer
(BASS)

• 12 racks CPUs, GPUs, network, disk, backup
• 256 16 180 CPU cores
• 180 NVidia G80 image analysis accelerators
• 46,080 floating-point cores, 250GB RAM
• 1 TB RAM (128GB 16x32GB )
• Infiniband 4X DDR connectivity
• 6 GigE connections (RENCI, BRIC, CISMM, )
• Red Hat Enterprise Linux 5
• Sun Grid Engine
• SAN

3
BASS Needs

• Already handled
• Hardware support
• OS support
• Installation and configuration
• Backups
• Needed
• Compile and run high-performance apps, libs
• Interface with RENCI, UIUC, IAM, BRIC, VL
• Web-based scientist access to codes
• Go to Supercomputing, bring back cool stuff

4
BASS Opportunities

• Experience
• Learn as much as you want about running a CPU
  CUDA GPU InfiniBand cluster
• Work with real scientists on real apps trying to
  solve CF, clotting disorders studying molecular
  docking,
• Networking
• Meet researchers across campus and around the
  country
• Participate in advisory board meetings with heads
  of centers and institutes
• Hobnob at the Supercomputing conference

5
(No Transcript)
6
Computer Visualization andTools for Science

• (Graphics Simulation folks, wake up!)
• (Non-Ph.D. intending folks can nap)
• Team Computer Science,
• Physics Astronomy,
• Applied Materials Sciences

7
Dragon Tamers Needed
8
NanoManipulator

• 2 Ph.D.s
• 39 Co-authored papers
• 7 trips to Europe and Asia
• Longest was 4 months
• Most recent last month
• 1 Commercial product

9
David Borland

• Flexible-Occlusion Rendering
• 1 Ph.D.
• 3 papers
• 1 patent application
• Licensed to Siemens

10
Summary of My Approach

• Teaching visualization (Comp 715)
• Multidisciplinary teams make custom tools
• Develop new techniques as needed
• Build entire systems as needed

Insights
Papers
Patents
Dissertations
Open Source
Commercial
11
Upcoming Challenges

• Blood Clots and Cell Cytoskeleton
• Image analysis Glowing webs ? models
• Optimization Fiber strain profile ? web shape
• Physical Simulation Nonlinear quasi-static sim
• Automated Microscopy
• Find me the interesting parts
• 3D image stitching
• GPU 120 Hz tracking of focus-changing spots

12
Roadmap

• Plan to hire 1-2 students in spring/summer
• Want to ensure long-term funding for you
• NSF proposal awarded in October
• Fellowship students welcome now
• 991 projects ready now

13
(No Transcript)
14
991 The Netflix Prize

• Image Analysis? Bioinformatics?
• Wake up!

15
Netflix Prize

• Increase by 10 ? 1M
• My plan ?
• Team registered
• Data downloaded
• Parser written and tested
• Data structures filled (need more RAM)
• Split N1 ways (1 share for UNC)
• Dont do it for the money

16
(No Transcript)
17
991 Your Favorite Topic Here

• Not just me, most professors
• Ask one, well point you to other likely
  candidates

18
Come See the Research!

• Graphics lab on 2nd floor, room 264
• 6-8 P.M.
• Meet the nanoGrad Students
• Graphics, Simulation, Image Analysis, Control
  Systems
• Talk with me about RA, Fellow, 991,

19
(No Transcript)
20
(No Transcript)
21
Extra slides below here
22
Find Great CollaboratorsSupport their Endeavors
23
Lessons

• Visualization based on goals, not techniques
• Avoid perceptual pitfalls
• Get over the rainbow
• Design specific tools, then generalize
• Experiment interface, not instrument interface
• Most definitely not computer interfaces
• Add novel devices as needed dont be a WIMP

24
Art vs. Science

• Science is what we understand well enough to
  explain to a computer.
• Art is everything else.
• Don Knuth

25
Tools to help ScientistsBuild Better Models

• Estimate models from images
• Adjust the models
• Estimate images from models
• Display model and experiment

26
Comparing Experiment and Model

• Scientist or image analysis estimates model
• Computer produces detailed simulation what would
  be the image for this model?
• Compare detailed simulation to experiment
• does the model predict actual image?
• Scientist or algorithm adjusts model
• is this a better fit?

27
The Right Tool for the Job

• All-in-one, not optimal for any
• Sears ShopSmith
• Computer (WIMP) interface
• Finely-tuned for the task
• Specific power tools table saw, lathe, router
• Automobile, airplane cockpit
• Bill Buxton Bow for violin
• What computers should be

28
(No Transcript)
29
Comparison in Model Space
What the scientist infers
The Data
Computed Centerline
30
Comparison in Experiment Space

• Simulated AFM scanner
• Scanned surface vs. model surface
• What should I be seeing?
• Ongoing collaboration with chemist Dorothy Erie
  to produce a model of what happens when proteins
  DNA are scanned with an AFM

31

• I am attempting to construct a mnemonic memory
  circuit using stone knives and bearskins Spock
• Star Trek City on the Edge of Forever