P. Oscar Boykin

Software patents are a tool for quashing independent programmers (commercial and otherwise). It seems that Europe may be on the verge of adopting the United State’s software patent mistake.

Many US programmers oppose software patents in the US and that includes me. It is sad to see Europe moving to embrace what I hope we here in the US can get rid of.

There are many resources about this issue:
FFII: Logic Patents in Europe
EFF’s Patent Busting Project

I am crossing my fingers that Europe will not make the same mistake that the US has.

I want an 802.11 keyboard and mouse. I have not seen anything like this. You might suggest I try bluetooth, however, I am a one-protocol-to-rule-them-all kind of guy. Imagine Wifi speakers which use the standard usb-audio drivers. Imagine a wifi headset: you could listen to music or do IP telephony far from the computer.

IONetworks makes some USB over IP products, but, these are general KVM switches and they don’t work with superior operating systems.

What I suggest is the following: someone standardize and submit to the IETF a protocol for USB over IP. Software would need to be integrated into the various operating systems to take these USB packets out of IP and put them into the USB stack. Also, security needs to be dealt with. But, (and I know this is hardcore geek stuff), but you could add components to computers that only had a regular 802.3 ethernet port. It works like this. You plug your computer into the LAN. On the LAN there is a wireless access port. Your keyboard and mouse talk to the wireless access port, and those packets get sent to the desktop. You could easily reconnect the keyboard and mouse to different devices, and there would be no need for a KVM switch. This does not end with mice, imagine wifi speakers or a wifi headset. I guess they would be much more popular than the mouse.

The major benefit here is that it would put downward price pressure on wifi chipsets and increase deployment of wifi networks. Who could imagine not having wifi in your home with all these applications?

The most challenging aspect is binding the device to the proper host in a secure fashion. I think that Zeroconf has a role to play here. Also, I would say that USB over IP should require IPSec: since keyboards are used to enter passwords, it would be a security nightmare to pass those over the network in the clear (if we instead did USB over TCP we could just use TLS (SSL). I don’t know enough about USB to see if TCP is more appropriate than UDP). A secure (but possibly costly) approach would be to embed a public/private key pair inside each device. Then, on a label printed on that device, print a hash of the public key. Finally, when binding the device to the host the user could verify that they binding was correct by comparing the hash of the public key the host was using to the one on the device (this rules out man in the middle attacks). Clearly, using one of these devices at boot time sounds like a tricky issue.

There are more issues to solve, but I think that unification of communication protocols around IP and ethernet sounds like a good idea.

Finally! Google gives us Google Scholar. It is similar in principle (but superior in practice) to Citeseer. We need more and better tools like this to accelerate the pace of research. Also, these sorts of tools will hopefully minimize the amount of duplication of scientific effort. It is amazing how many problems are solved over and over again simply because researchers are unaware of previous solutions.

New preprint on the Arxiv: “Threshold Error Penalty for Fault Tolerant Computation with Nearest Neighbour Communication”. Catchy title huh? This is actually a nice paper in my opinion. It shows how to do fault tolerant quantum computing with a 1-d array of qubits. We find various error thresholds depending on different assumptions, the range is from 10^(-4) to 10^(-7) (probability of gate error per time step). These can be improved somewhat.

A nice compliment to this work would be a project to carefully study the physics of some proposed quantum computing models and see what the attainable error rates are likely to be. Is 10^(-7) totally out of the question? Can we find some cases where bit error rates will be higher than that even in principle? It would be interesting to do some analysis akin to Asher Peres, Petra F. Scudo and Daniel R. Terno’s “Quantum entropy and special relativity” to find cases were the error rates will be above threshold even in principle.

My co-author Jesse Bridgewater just submitted our new paper “Balanced Overlay Networks: Decentralized Load Balancing via Self-Organized Random Networks” to the Arxiv. This paper is an improved version of the algorithm from the paper “A Statistical Mechanical Load Balancer for the Web”. In the current paper, we use a much more effective algorithm, however it is one for which we have not yet analytically solved the dynamics.

This new algorithm works basically the same way as the previous: nodes accept jobs from other nodes, and they keep an in-degree (on a directed graph) proportional to their unused resources. In the current paper, we show that by taking a short random walk on the graph, and sending a job to the least loaded node on the walk, we get an almost perfectly balanced graph: the degree distribution is almost a delta function.

This could be very useful for web mirrors of such software as Debian. If there was a small Apache module which kept this network, Apache servers could easily redirect clients to nodes on the network which are less loaded. This would make for near optimal usage of mirror bandwidth.

I have used the same email address for the past 9 years. This email address collects something like 1500-2000 spam messages per week. Thanks to tools like CRM114 and Spamassassin, I keep most of it at bay (better than 99.5% I am never bothered with). However, most users do not have such spam filtering: it is still too hard.

This is one area where I am quite critical of “Industry Leaders” such as Microsoft. According to Pew, Spam and the authenticity of email is one of the biggest problems facing internet users today. If a couple of open source programmers (such as those who have written CRM114 and Spamassassin) can make my spam problems largely disappear, why hasn’t Microsoft integrated spam filtering into the OS to defeat IM spam, Email spam, Blog spam, and so on? If there was an effective engine built into Windows (which many people use) spam would be on the ropes. Currently only 50 in one million people respond to spam, but that is enough to keep it profitable. If MS installed spam filters on 95% of the world’s desktops, we may see that number fall from 50 to 3 people in a million, which may put many spammers out of business (due to not making enough money).

As a (small time) blogger, I also get many spam posts. I am trying to deal with it using a Wordpress Spam plugin, we will see how it goes.

The bottom line is that there are 6 billion people. Suppose if even 1/1000 of them want to sell you something (which I guess is an underestimate), we can expect 6 million potential spammers. If communication is free, we can expect more than 6 million offers. Obviously, this would mean all communcation is totally dominated by spam. Reminds me of the Minority Report. Spam is a very serious threat to the value of communication networks.