Lots of software and software-engineering discussion in this blog today.

MOSFIRE data reduction pipeline

Worked with Andreas Faisst on a nasty mosfire data reduction bug. We engaged in some pair programming which turned out to be quite useful for me. I hope that in the future Andreas will be able to help me in the future!

Reed Riddle's Talk

Reed gave a nice talk about robotic astronomy. The key point is an essential one. It requires deep though to design something that is efficent, maintainable, and flexible. It is challenging to make a system behave well. Software often grows organically, sometimes organic growth works. Reed argues that best performance comes from good design -- though this is never demonstrated. The talk was architected to demonstrate both sides of the coin, first with TMT second with robo ao.

TMT Site Testing

Pre-selection of five candidate sites based on satellite studies of cloud cover and water vapor. Used identical equipment with emphasis on proper cross-calibration validity.

TMT Candidate sites:
Tolar, Armazones [VLT], Tolonchar, San Pedro, and Maunea Kea.

Out of a suite of 12 instruments there was one seeing monitor and then some kind of turbulence or environmental monitor. The sites are environmentally tough so software has to be robust. The site-testing software was significant with 7 operating systems plus 12 scripting and programming languages. All of this grew organically.

The organic growth meant that there was no way to design the system. As a result, the operations were sequential requiring blocked state. But as a result there was no coordination between the instruments and efficiency is low with something like 30% science time. It was also to difficult to maintain this system.

Learned that there is no single right way to create a system. Early design decision affect the operational efficiency. No initial planning for software. It was hard to find time to make the system robust. Ultimate lesson: temporary fixes become permanent fixtures. Perfect is the enemy of good enough.


Fully automatic adaptive-optics system. Lots of good details at the robo-ao website. End result is that seeing-limited images of 1" are converted into fraction-of-an-arcsecond images produced by Robo AO.

'Half the resolving power of HST at less than a ten-thousandths of the cost'

Robotic ao operations usually require a large staff. The vis AO at magellan has a team of engineers and staff monitoring the AO system.

Robo-ao uses a modula control software syste. Robotic system, watchdog, system monitor, data system, and status system. Each subsystem is a demon and the demon has inherited a variety of code from one and other.

Performance of robo-ao is impressive! Over half the observing time is science time.