Life as a Physicist

Only one speaker has managed to generate spontaneous clapping in the middle of a talk here at CHEP. It was Axmark, during his talk on MySQL. The comment was something like “We should get rid of software patents.”

I suppose I don’t fully understand how this works. The patent system was built to offer protection to inventors and, at the same time, release information to the public so all could benefit. If you came up with a new widget you could patent it. That means no one else could make it without your permission. But at the same time everyone could check out how you did it and perhaps apply some of the same techniques to their unrelated problem. An attempt to both publicize the intellectual information and at the same time protect investments.

Software is kind-a funny, however. A lot of the patents that are around seem quite silly. Remember the one-click patent? I would be quite happy to see patents like that to the way of the toilet.

But what about a company that puts 4 or 5 years into inventing a new algorithm? Coming from the science end of things, I think they should publish their work so that all can use and improve upon. But if that publication means they right-away looses a good deal of their money-making power in the market.

I’m sympathetic to the arguments that there are other revenue models. Open source has shown us a proven one: support and improvements for those customers that pay them. That is a good model, and it supports improvements to the product. If we eliminate software patents does that mean that everyone has to have a company that runs this way?

On the other hand, software isn’t like devices which the patent system was designed to protect. You can take apart something mechanical, reverse engineer it, and see how it works. With software that is much more difficult (indeed, perhaps illegal in the US what with the DMCA). Under that model you don’t need software patents at all. Only the public in general never gets to learn about the cool new algorithm the company has invented. It also makes it harder to see if a company has stolen code that is protected by a patent.

My conclusion is we need software patents, but not as they are structured now. How should they be fixed? I have no idea.

A friend whose daughter works in the US Patent office said that she was in favor of getting rid of software patents all together. The biggest problem is they have no way to search to see if something had been done already (no prior art). Her claim (if I understood him correctly) is that a lot of the problems stem from this difficulty.

I’m not sure exactly why, but everyone loves to write a monitoring system.

Monitoring is key to the operation of any DAQ system. It is the only way to monitor the health of the system in real-time. Plots, diagnostics — the long term stability of the system — all depend on a good monitoring system.

In a large experiment many components of the detector are built and tested in isolation. Even at the testing level monitoring is developed. Usually one of the last things (hey — get the hardware working or spend time on the monitoring — choice is obvious). So, it is frequently done at the last minute and done quickly, and in each detector sub group. Then it comes time to integrate it together with the other bits of the detector. But at that point your particular monitor system is integrated with all your code… so… we end up with multiple monitoring systems.

We did this in DZERO. CDF has done it. And now, I see here at CHEP, ATLAS and CMS and probably others are going to do it (46, 58, 218, 244, 269, 274, 283, 308, 406). And as I scaned the web page to make that list, I realized that perhaps the same thing is happening with the GRID — there are lots of them sprouting up as well.

The trick comes when things settle down to running. Now you need to aggregate all the monitor information to build dash-board like displays. And you need monitor information from multiple monitor systems. Ops. Programming nightmare.

We learned a bunch of lessons when we did this at DZERO (we’ve talked about our monitoring system at previous CHEP’s):

• Keep it light weight. A simple TCP/IP wire protocol is more than enough. After all, there is very little control information: mostly just data. Kill CORBA now!
• Keep it extensible. Any developer should be able to add data to the system without having to modify central components. Further, they should be able to add complex monitor items that aren’t generated 99% of the time (i.e. debugging information).
• Keep it open. Make it easy for anyone to access the monitor data. It is amazing what people, not at all part of the daq system, will invent at 2 in the morning if it is easy to get to.
• Try to unify the systems. Multiple monitor systems will plague the experiment for years. Kill them early and often.

I fear it is already too late for ATLAS and CMS.

UPDATE: See the comments from people who are actually working on these systems (as opposed to me looking in from the outside).

Liz gave one of the CHEP plenary talks on software frameworks. These frameworks are used to reconstruct our data and also to analyze it (i.e. extract the physics). The framework is amazingly important software: everyone in the experiment touches it. Usually one or two experts write it using all the features of C++ they can.

I had two issues with this review of framework history, titled “Event Processing Frameworks; A Technical and Social Challenge.” First, she correctly makes the point that the physicist will vote with their feet: if something is too complex or too slow for them to use, they will move to something else. She then talks about how getting physicists to use the features of the framework correctly is like herding cats (and played that crazy video). Took me a minute to get over being called a cat…

Second (related to the first) Liz said that you can measure the success of a framework by two criteria: how well it runs reconstruction and how well it runs physics analysis. The two tasks a very different. Reconstruction is big iron: you need to know exactly where you crashed (if you do), you need to be able to track every bit of data, you want to make sure you know exactly what code was run. It often takes 6 months to put together a new reconstruction executable. Analysis, on the other hand, is fast and dirty. You make a change in your code and you want to see the change in the plot in 10 seconds. The former requires lots of infrastructure support. The later really wants a lightweight framework.

It isn’t obvious that they should be the same. In fact, looking at the landscape of physics analysis frameworks — in the end, almost no one uses the same framework for reconstruction and analysis. In DZERO, for example, we use one framework for reconstruction and then use the ROOT tool for analysis. From those I’ve talked to this is typical. ATLAS is still hoping their framework will be used for both, but I think the days are numbered. We will all be using (gasp!) ROOT.

Both Rene and I asked questions along this line (as typical, he was more blunt and to-the-point than I was). I think Liz still believes that a framework should do both.

[There will probably be a few technical posts while I attend CHEP 2006. Sorry!]

One of Time Magazine’s cover stories addresses the science environment in the US.

The Federal Government, beset by deficits for most of the past three decades, has steadily been cutting back on investment in research and development. Corporations, under increasing pressure from their stockholders for quick profits, have been doing the same and focusing on short-term products. And the quality of education in math and science in elementary and high schools has plummeted, leading to a drop in the number of students majoring in technical fields in college and graduate school. In the past, hungry immigrants looking for America’s prestigious Ph.D.s made up for that decline in the U.S. science and engineering labor force. Now if they come to America for Ph.D.s, students often return with them to gleaming labs in their homelands.

This is a good nutshell description of what is happening. One of the biggest problems we have is convincing people that drops in research funding, focusing on short term results, etc., really is bad for the US. Sure it is obvious, but can you find a direct link? It is like global warming: science innovation comes from many sources. If you watch over time you’ll see the (science) environment slowly deteriorate. By the time you can prove the cause, it is too late.

This is why I’m hopeful after the President’s budget was released. Apparently motivated by a NAS report, a number of senators and now the white house have responded. Lets hope it makes it!

I burry my head to write my talk for a few days and our Veep shoots a lawyer!? What the heck!

Next: get the btagging working!

The first version of this post started with my impressions of Mumbai. So did the second version. Both of those were lost to the bit-bucket. I’d really like to whine about the internet, but I’ll hold off for another post.

My seat mate on the 8 hour DC-10 flight from Amserdam to Mumbai told me the first thing I’d notice were the sheer number of people. When he first moved to New York city he used to go around telling his friends how much empty space New York City had. Now that I’ve been here for 12 hours I see what he means!

At the gates of the airport immigration there must have been close to 1000 people waiting for friends/relatives to arrive. Only the metal gates and railings gave us room to actually get out of the airport.

It was 1am when we arrived. As I walked around searching for our ride to the hotel and an ATM machine one thing I noticed was the sheer number of people who were sleeping on the ground! They were everywhere, and totally out despite the noise of those other 1000 waiting for friends to arrive, the beeping of horns from the parking-lot traffic jam… Wow.

The conference has hired drivers for all 480 of us to get us from the airport to the hotel. I’ve never been to a conference that did that before. Now that I’ve read the registration packet I notice that they have also scheduled dinners on almost all the nights for us. I can’t help but wonder if that isn’t to insulate us from the rest of the city.

The drive from the airport to the hotel was something else. I’ve never felt the need to wear a seat belt in the back seat of a taxi. Before this. Someone could make a lot of money if they sold a device in India that would flash the lights repeatedly and honk the horn. Lane markers? Not sure why they waste money on painting those things on the roads. Traffic lights and signs? To call them even advisory would be a stretch.

But I’m here. And it is pretty spectacular. This is very different than going to a European/Western city. Just walking down the street you know you are in India.

I will upload pictures when I figure out how the internet is going to work.

Last Tuesday’s New York Times has a great article on digital photographs submitted to scientific journals. It used to be pictures were submitted as pictures — photographs. But a few years ago they started requiring, along with every other journal, started requiring fully digital submissions. There was a surprise waiting.

They started checking illustrations in accepted manuscripts by running them through Photoshop and adjusting the controls to see if new features appeared. This is the check that has shown a quarter of accepted manuscripts violate the journal’s guidelines.

Wow! What is a violation?

Dr. Rossner realized that some authors had yielded to the temptation of Photoshop’s image-changing tools to misrepresent the original data.

In some instances, he found, authors would remove bands from a gel, a test for showing what proteins are present in an experiment. Sometimes a row of bands would be duplicated and presented as the controls for a second experiment. Sometimes the background would be cleaned up, with Photoshop’s rubber stamp or clone stamp tool, to make it prettier.

A lot of things they talk about are pretty harmless in the sense they probably didn’t  change the reported scientific outcomes. And sometimes you do want to photo shop things to emphasize the point, I suppose. But the journal put down its rules for a reason.

By the way, the way they discover these modifications is pretty cool — and simple. They change the contrast up and down and alterations would show up as boarders. Or enlarge the image and notice that the pixels were different sizes.