I like to avoid using the lock statement in C# for short bits of code that don’t block. This is partly based on an only loosely supported idea that Interlocked operations are faster for simple code. I often use the concept of a protected immutable state object as an inner class of my object. Then, my objects have only one mutable member variable if mutability seems necessary. To update the state, I use code like this:

State state = _state;
State old_state;
State new_state;
do {
old_state = state;
new_state = SomeFunction(old_state, arg1, arg2, arg3);
state = Interlocked.CompareExchange<State>(ref _state, new_state, old_state);
}
while( state != old_state);


The above code works logically as: read the state, compute the new state based on the old state, and then, if the state is still the same as it was when we did our computation update it. CompareExchange returns the value of _state when it is called, so by looking at the result we can see if we were successful, or if we need to try again.

I could encapsulate this pattern as:


class Mutable<StateT> {
protected StateT _state;
public StateT State { get { return _state; } }
//Use the delegate how_to_update to map the
//old state onto the new state and return the new state
public StateT Update(System.Converter<StateT, StateT> how_to_update) {
StateT state = _state;
StateT old_state;
StateT new_state;
do {
old_state = state;
new_state = how_to_update(old_state);
state = Interlocked.CompareExchange<StateT>(ref _state, new_state, old_state);
}
while( state != old_state);
return new_state;
}
}

Since delegates can (or at least once did) have performance issues, it doesn’t make a lot of sense to do the above for highly performance critical code, but of course, one should only obsess about that if benchmarking shows that a particular usage is causing problems.

The above statement may call into question the whole approach: why not just update _state in a lock? One reason is that the above code can never deadlock since how_to_update is not holding a lock when it is being called. Of course, you can do the same logic as CompareExchange with a lock, but at that point, why not just CompareExchange?

I’m trying something new this semester. I’m going to actually try use the wiki as it was intended. There will be a page for each lecture. Students will take turns making notes and posting those as wiki-fied pages. Hopefully this will make it easier for students to follow the course, and generate more valuable content for all.

See you in the classroom!

Welcome to the fifth installment of Academic School Years, featuring me, Professor Boykin. In this exciting episode, I’ll be teaching EEL 4834, Programming for Engineers. The 4834 course web page is on my wiki.

As the above XKCD comic depicts, Python is more fun than perhaps any programming language has any right to be. Nevertheless, we will not question that fact, but accept it, and master the wonderful language.

I often get automatic email messages with the reply-to address set to some kind of “noreply”. This is lame. For instance, Facebook sends notifications using such a noreply reply-to address. To me, this seems like lazy programming. If someone sends me a message on facebook, if I reply to the notification I am clearly intending to reply to the message. Facebook should parse that response and send it on to the original sender.

The same is true of twitter direct messages.

Systems developers should make their notification channels (IM, Email, SMS) two-way and not just one-way conduits. Not doing so represents a failure of design.

You can see the mercurial repository here, or download the script itself here.

The SocialVPN project uses social networking to build P2P VPNs. It makes it trivial to communicate, play games, share music, etc. with anyone in your social network with any IP-compatible software. You can find the slides here. Below, you will see a photo of me giving the talk:

In addition, a related work: “Improving Peer Connectivity in Wide-area Overlays of Virtual Workstations” by Arijit Ganguly, David Wolinsky, and Renato Figueiredo. and myself was presented at HPDC 2008 by Arijit Ganguly. This work also won the best paper award.

Congratulations to my co-authors!

1. I have installed Rockbox on my iPod.
2. Amazon has released a non-free GNU/Linux client for their MP3 store.

With Rockbox, I’m using free software when I use my iPod, but there are other benefits than ideological purity. Rockbox supports many players, so I can easily buy a new mp3 player and get the same features I’m used to with my current device. The main feature I like with Rockbox is that the device maintains its own database. To install music on it, I only need to copy the files. My Ipod appears as a normal USB disk. This means I can use rsync to automatically keep my iPod as a mirror copy of all my music. When I go to a new machine, I can use rsync to copy all my music off just as easily. In this way, my Rockbox-powered iPod becomes an exact mirror of my music collection. Since I’m using Emusic (and now Amazon) to buy mp3s, keeping backups is very important to me.

The second question is the acceptability of Amazon’s downloader program. It is non-free software, but I’ve allowed it. The reasoning is the following: it enables a single transaction. When that transaction is over, it gets out of the way. I don’t need to use the program ever again. If in the future, Amazon does something to violate my rights, I just uninstall the program and keep using all the old mp3s. I don’t really see this as materially different from using their web page (which has javascript on it and as such executes code on my computer).

As a practical matter, I will only buy mp3 albums from Amazon when it is cheaper than buying the CD used (including shipping).

This is the first time I did not get a personal record (PR) at a race. This is due to training mistakes on my part. Like many athletes, I have a hard time admitting when I am injured. I got a minor injury prior to my last big race (Micanopy half-marathon). I was pretty happy with Micanopy, but it greatly exacerbated my piriformis syndrome. I did take a lot of time off running after Micanopy. I was starting to feel a lot better, so I figured, why not race 9.3 miles. Turns out, that was a dumb idea. Here is a plot of my pace as a function of distance:

You may note that around 6 miles my pace starts to slip, and after 7 miles it really drops. At that point in time, I was thinking about walking because I was in some pain. Like many a foolish runner, I wasn’t brave enough to put my health before the embarrassment of not finishing a race. Fortunately, it doesn’t seem to have caused major problems, but I definitely start to feel some discomfort when I get over 5 miles or so now.

Now, I’m doing everything I can to recover. You name it! Stretching, strength training, cross training on the bike, stairs, swimming, self massage, active release therapeutic massage, yoga, and even…. acupuncture. I’m looking for anything at this point: placebo effect, real effect, who cares: I’m just looking for effect. So far, things seem to be going well, but I’m definitely going to focus on shorter distance running until this heals up.

My current theory is that I’ve injured a ligament near the piriformis and also injured the sciatic nerve. Both ligament and nerve injuries just take time to heal, and clearly, I’ve not given them enough time yet. Like almost all runners (or even all athletes), it’s really hard for me to take it easy while I recover, but I try to tell myself that everything I do to heal is really part of my running. When I’m swimming, I’m running. When I’m stretching, I’m running. When I’m doing yoga, I’m running. All is one.

Two somewhat related notes. I got the data out of my garmin 305 using Garmintools, a free software package for interfacing with the Garmin 305. I recommend it if you run GNU/Linux and have a 305. Secondly, I did link to the photographer Richard Ritari’s site, but I can’t even view it properly because his site now uses flash to view simple pictures. I guess this is due to some quasi DRM-like idea of trying to make it hard to keep thumbnails of his images. I bought one image from him in the past, however, I doubt I would again. In the past, the digital download cost 40 dollars. It seems he is no longer even offering them, instead you must buy a 4×6 or 5×7. Just a note to photographers out there: it’s 2008, I’m not buying prints anymore. Charge me 10 dollars for the digital download (of pictures he has already taken and stored), and I will buy every picture he takes of me. In the mean time, he’ll be getting zero dollars from me.

One thing they didn’t mention which I think bears repeating is why spam filtering is so hard. Here is what I said:

Spam filtering is an adversarial problem. You want your email spam free, but
spammers want you to read their messages. Most technological problems don’t
have this aspect. The spam problem is more similar to automated crime
fighting than building a better search engine, for instance. As text spam
filters have increased in accuracy we’ve seen spammers move to image (jpg and
gif) and pdf based spam.

Spam classification is hard. If you hired a human to do the job, you may not
get better results. Certainly some messages are easier for a human to
classify than a computer, but the reverse is also true. Bill Yerazunis,
author of the spam classifier CRM114, estimated his own classification
accuracy as 99.84%, which was worse than his software did:

http://crm114.sourceforge.net/wiki/doku.php?id=news

I suppose I was asked due to my research in spam filtering based on social email networks, work which has been in the news before (Slashdot, New Scientist, Nature News).

I extend a very warm welcome back to school to our fine University of Florida students. This semester, some of you will have the dubious honor and privilege to learn Information Theory in a course conducted by myself. This subject is dear to my heart and I find the material to be quite beautiful. I look forward to sharing my enthusiasm and knowledge of the material with you. To put in a vernacular more accessible to today’s youth: I’m totally pumped about this semester!

Please take a look at the website for this Information Theory course. There is also a Google Group for the course. Please join the group and introduce yourself.

See you in the classroom!