Concurrent Python using transactional memory

Speaker: Fuad Tabba

"Parallelism is hard"

• Figure out the parallelism in the application.
• Figure out the required synchronisation. How do you protect the critical sections? Race conditions, deadlocks, livelocks. Locks take away the parallelism you had hoped to achieve.
• Locks have inherent overhead.

Python uses one "Global Interpreter Lock" to protect shared resources.

Transactional memory: Atomicity, Consistency, Isolation, Durability.

The use of transactional memory is to abstract away critical sections so that parallel programs can become easier to write.

In the Sun Rock Processor (which has limited hardware support for transactional memory), there are two caches:

• L1 Cache tracks memory locations that have been read and written to.
• Write buffer stores tentative writes (uncommitted transactions).

Cache coherence protocol is used to detect transaction conflicts. The physical size of the caches limit the size of the transactions. When a transaction is committed, the new values have to be propagated to each processor. Transactions can abort/fail for unspecified reasons.

SGI

• Simulation examples (Hardship)
• Bandwidth of copper wire doesn't keep up with bandwidth of processors.
• Latency of conducts not improving.
• Fibre optics not very fast. Only air core; but increases available bandwidth.
• mpi_reduce >> mpi_waitall, mpi_reduce, mpi_barrier, mpi_recv.
• Ethernet chip to offload communication.
• Memory chip speeds have not increased.
• Bandwidth/Latency.
• Linux -> Computer frontend -> Matrix of computing resources.
  

Experience with and potential of hardware transactional memory

• Rock processor
• Everyone needs to do concurrent programming. Not just OS or VM developers.
• Essence of TM: ability to access multiple memory locations in an atomic transaction, without specifying how atomicity is achieved.
  
• Using one lock: not scalable but mimicks transactions. Parallel critical sections.
• Finer grained locking.
• Best effort hardware transactional memory - Hardware can abort. Problems raised to the software level.
• Abort feedback important.
• Speculative haredware features gave subtle bugs.
• Lock elision.
• Doubled ended queues.

Bill Payne

Friends, Family and Fools (3x investment levels)

• Unsophisticated
• 1 -2 times a lifetime
• Passive
• Gift?

Angels (1x investment levels)

• Time and money
• Active
• Investing in entrepreneur
• Accredited
• Savvy

Venture Capitalist (1x investment levels)

• Institutional money
• Large portfolios
• Invest in company
• Limited partnership

Capital lifecycle













The funding gap

• Between Angel and VC.
• $1 US <--> $4 million US.
• Entrepreneurs: Go where investors are.

NDA

• Not motivated to steal technology.
• Most Angels and VC will nto sign.
• Write non-confidential business plans.

Friends, Family & Fools

• Debt, equity or gift?

Angels

• Accredited investors
• Exited entrepreneurs
• "Mad Money": Money set aside to lose
• Range of roles
• Lead investor
• Investor/advisor
• Passive investor
  

Motivation

• Venture capitalists make money
• Return on Investment
• Staying involved

Business plans

• Elevator pitch: Verbal. Attract interest. Don't close.
• PowerPoint: Verbal. 10 pages, 20 min, > 30pt font.
• Executive summary: Written. Attract interest. Don't close.
• Full Business Plan: Written. Validation score card. Write full plan first.

Evaluating deals

• Management team > Opportunity > IP > Funding
• Investing in the business, investing in you.

Fundable management teams

• Vertical experience
• Balanced / Complete team
• Coachable
• Leadership
• CEO experience

Size of opportunity

• High gross margin
• Scalability
• Large niche market
• Unfair competitive advantage
• Ready for customers

Angel investing process

• Pre-screening
• Screening
• Due Diligence: Also check investors
• Investment presentation
• Follow up
• Closing
• 1 in 72 who apply receive investment

Five mistakes to avoid

1. Don't take dumb money. Someone who is passive or does not help you.
2. Don't elaborate in describing your product / technology.
3. Don't overestimate market size.
4. Don't ball park addressable market size as a percentage.
5. Don't press the first mover advantage.

Finding money is time consuming.

Prefer to see a company with entrepreneurs with majority control of the company. not interested in taking control of the company. Prefer to keep entrepreneurs motivated.

Lack of execution on the part of the entrepreneurs.

If the total investment is more than $10 million US then company would probably not succeed.

Self-Sabotage

Patterns of behaviour

• Overcommitting
• Never saying no
• Getting distracted
• Perfectionism (measuring against)

Impostures syndrome

• One step away from being found out as a fraud
• Keep being an impostor
• Others are also impostors

Saying no

• Automatic Negative Thoughts (ANTs)
• More Accurate Thoughts (MATHs)
• ANTs ... and so ... MATHs

It's the thought that counts

• Depends what you do with the thoughts.
• Event -> Beliefs (intervening probabilities) -> Feelings

Twisted thinking (wrong conclusions -> wrong actions)

• All or nothing
• Over generalisation
• Mental filter
• Discounting the positives
• Jumping to conclusions
• Magnification
• Emotional reasoning (feelings as facts)
• Shoulds (assuming how things should be)
• Labelling
• Personalisation and blame

Procrastination

• Avoidance strategies
• Action -> Motivation -> Action -> More action
• Break into smaller pieces
• Procrastinators: Leaders of Tomorrow

Time management

• Velcro fingers
• The three D's
• Do it
• Diarise
• Ditch it
• Pareto principle (80/20)
• 20% of the work leads to 80% of the output

Circles of influence

• Core: Can control
• Inner circle: Can influence
• Outer circle: Can't control

Writing is not recording, it is thinking

Lunar Surface Systems: International and Technology needs

Dr. Michael R. Lowry

• James Webb Telescope (2014): Successor to Hubble Telescope.
• Kepler Telescope (2009): JPL + NASA. Has photometer. Searching Earth sized planets. Higher signal to noise ratio than ground based telescopes.
• LCROSS (October 9 2009): Lunar CRater Observation and Sensing Satellite. Threw up 90 litres of water into the air. Recoverable water. Anomaly: Communications black out. IMU star tracker failover. Ping pong between control deadband. Lack of sensor resolution. Insufficient full failure mode analysis. Software patch uploaded.
• Aircraft like design of Low Earth Orbit spacecrafts. Longest runway. $1.5 billion amortised cost for every use.

• Constellation: Solid rocket booster carrying occupants. Altair Lunar Lander (heavier). Pad abort test (new capsule).
• Augustine Commission: Budgetary considerations. Mars exploration. Launch alternatives. "Flexible path".
• Interplanetary Flotsam and Jetsam: 4.5 billion year old rock from Mars. Fossil evidence. Primitive Martian life.

• Getting to Mars
• Preliminary Lunar Surface System. International guidance.
• South poles are good. Always sunny. Use solar panels perpendicular to the ground.
• Space transportation: Rocket equation. Carry mass (fuel) to and from Mars/Earth. Launch window, Flexible Path, Aero braking, Production of fuel remotely, Ion propulsion.
• Rover: Radiation protection, dust mitigation, guidance matching terrain maps, teleoperated navigation.
• Communication: Space certification, disruption, QoS, reconfigurable networks.
• High efficiency, high recovery ECLSS (Environmental Control and Support System).
• Light weight structures.
• Smart home systems: iPhone application.
• Planning done at Mars and only call back to Earth when help is needed.
• Safety certification: Class A mission standards.
• Plug and play certification of open architectures.
• Interoperability.
• Extend human civilisations

• Ancient systems for radiation resistance.
• Mars: A death cemetery for robotic explorations.
• Contracting out parts of missions.

Michael.R.Lowry@nasa.gov

Part 4 physics presentations

Missed the first presentation.

Anna Yang - Mechanical model of the breast

• Aid clinicians
• Cancer research
• Segmentation of regions from MR images
• Three interests: Air, skin, fat
• Using active contour models to find a 3D model from a series of pictures.

Matthew Walbran - All fibre Coherent Anti-stokes Raman Scattering

• CARS for spectroscopy
• Multiplex CARS to cut down tuning time.
• Fibre being an optical fibre.

Ihab Ramadaan - Investigating the use of "colour" spectral CT

• Imaging modality
• Distinguishing different oragans/tissues
• Current techiques use markers but triple xray doses and time. Also dual energy scans but again double the time and xray. Also spectral CT works like RGB
• LHC = Large Hadron Collider
• Testing spectral CT and acquisition developed in collaboration with the university of Canterbury.
• Stopping power determined by the sensor layer.

Yukti Srivastava - Construction of an EDFA for use in an adavance lab

Speaker is crazy with the laser pointer. She needs a light saber!

• Amplification of light in long fibres.
• Connectorised components of the lab equipment.

Pratik Raval - LED as replacements to traditional lighting.

• Light is confined in LEDs because of internal reflection. Angle of illumination is small as well.
• Thermal management of LEDs due to thermal output over a small component size.

Poor guy. Not sure what his contributions were.

Jason Kuo - Bacterial tagging with optode

• Using labview and a single fibre probe
• Ended up a two fibre probe to compact coupling issues causing an irregular peak.

Amy Lin - Construction of a femtosecond Kerr shutter

• Need to combact "chirp" in super-continuum production
• Super-continuum source
• Optical delay line
• Polarizer
• Kerr medium
• Orthoginally, a femtosecond pulse source
• Analyzer

Kaidi Liang - Super-continuum generation

• Making tapers for super-continuum generation
• Third order effects most prominent in super-continuum generation
• Taper = heating a fibre and stretching under control.

Terminology issue: core

Neil Alan Campbell - Determining iron loading using MRI T2* decay curves

• Iron loading is the amount of iron in tissue.
• MRI for determining iron loading without having to take biopsies.