Juggling Molecules with Linux

An anonymous reader writes "This article at LinuxDevices.com describes an interesting project at the University of Vermont in which researchers use real-time Linux to build a laser trap that manipulates individual molecules by means of a computer-controlled laser beam. The project makes use of RTLinux, a real-time enhanced version of Linux that allows the system to process interrupts every 50 microsecond, sample new data, and timeshare the laser beam position. 'If the computer failed to respond, for even a millisecond, then we would drop the balls,' explained one of the researchers. Gives a whole new meaning to BSOD, eh?"

Victor Yodaiken if a Patenting Pest

[Anonymous Coward]

Victor Yodaiken (who wrote TFA) is the clown who patented his technique for implementing RTLinux. I much prefer RTAI for real-time linux, both because it is IMO a superior implementation, a better license, and it doesn't give support or credibility to Yodaiken.

Re:Realtime Linux on the desktop.

[akozakie]

That's also the main selling point for desktop dual-core chips - without a realtime OS the second processor keeps the GUI running smoothly even under load. I sometimes use my friend's old dual Pentium 300, and under load it actually feels marginally faster than my (way faster) single chip Athlon. If the price, noise and energy consumption weren't so high I'd buy one for home use - and dual core may be the solution to those problems (plus, dual-core notebooks are a possibility, dual-chip - hardly). I also worked with QNX - nice...

In short - I want realtime Linux on my desktop NOW! I wish I wasn't that lazy and would actually do something about it...

QNX - for really low latency

[Animats]

Here's what comparable numbers look like for QNX: QNX [ddjembedded.com].

For a 200MHz Pentium (this is an old review), the testers tried sending one billion interrupts with a latency check. When they required 8 microsecond latency, they missed one interrupt in a billion. When they only needed 10ms latency, they didn't lose any.

Comparable figures are available for various real-time Linux systems. [linuxdevices.com] Note that these figures are for a 650MHz CPU. The times are slightly better than for QNX, but the CPU is 3x faster.

Bear in mind that "RTLinux" programs aren't running under Linux. They're running below Linux. They can't make most system calls, for example. QNX programs are ordinary programs, and can make system calls.

The Linux 2.6 kernel isn't bad, though. Running real-time with millisecond response as high-priority Linux threads can actually work in 2.6. In 2.4, no way. You have to be very careful not to load any high-latency drivers, though.

Re:Laser Traps

[ZagNuts]

I was under the impression that the whole idea of a laser trap is that you CAN'T drop the ball. Small particles get trapped in the beam due to photon pressure, if the particle shifts away from the center of the beam, it automatically is recentered. Then you can move the beam to manipulate the particle which is attached to a molecule. They use these to fold and unfold proteins, lipid layers, DNA, etc. I mean, it's great that they're using a realtime kernel, but they really shouldn't NEED it

From the article:
"In other experiments, researchers use the laser beam to apply minute forces to the beads and to the connected myosin molecule."

They need real time feedback in order to compensate for motion caused by applied forces that might push the molecule out of the focal point of the laser.

They may also be using realtime feedback to continually focus the laser since very small variations can cause changes large enough to drop the molecule (although I believe many of these systems use purely electronic/optical means to do this).

Re:Why do we have this statement?

[monopole]

If you ever had to perform real time processing using Microsoft Windows you would regard the comment as kind. Windows employs a constant blizzard of interupts which makes response times unpredictable at this scale.

Actually the BSOD is the least of the problems, with lags and leads being the primary problem.

Re:delay tolerance?

[ClosedSource]

Yes 1 millisecond is pretty slow. On the Atari 2600 we used to meet a .84 microsecond timing requirement with a 1.19 MHz processor.

In any case I don't think many people on Slashdot understand that tough, classical real-time software can't really run on a PC (or Pentium processors for that matter) no matter what OS is used.

The key to real-time software isn't speed, it's deterministic timing. Once you have a cache involved, it's pretty much game over. Unless, of course, your timing requirements are several orders of magnitude slower than the time it takes the processor to execute an instruction. In that case the non-deterministic behavior may be swallowed up by the large gaps between real-time events.

Nevertheless there may still be the possiblity of memory management accessing the disk and blowing your timing away.