P. Oscar Boykin

I am back from the SpamConference. Slashdot has a thread on the SpamConference also. I had a lot fun, and just missed the bad weather.

Of the talks, I was particularly interested in (listed in order of presentation):

• a talk presented by Shalendra Chhabra on a Unified model of spam filtration.
• a talk by Jonathan A. Zdziarski, the author of DSPAM, on bayesian noise reduction
• a talk by Richard Segal, of IBM Research, on Classifier Aggregation. They show a nice way to combine multiple filters and show empirical results. A paper on the subject also appeared in the Conference on Email and Anti-spam.
• Matthew Prince from Unspam presented a talk on Project HoneyPot. This looks pretty interesting. They are collecting a ton of data and they say they will share it with researchers.

Finally, I got to meet many interesting people, including Meng Wong, founder of Pobox.com (a service I have used since mid 1995). Recently he has been helping the world with his work on SPF, Bill Yerazunis author of CRM114, as well as all of the speakers.

I just finished my talk at SpamConference 2005. Look for this work to appear in IEEE Computer soon. In the mean time, find my slides here, and the older preprint on the arxiv (it does not include results about bootstrapping).

Someone asked about my figures from the slide. You can find them here.

I am sad to say that for the moment, I have disabled comments to the blog.
Due to the low rate at which legitimate comments appear, I doubt many will be troubled by this news. On the other hand, I get around 10 comment spam posts each day. Sometimes they are prevented by Kitten’s Spaminator, but not always.

Recently I have seen some (as far as I can tell) useless comment spam posts. They all basically say “Very useful site….blah blah blah”, but they don’t include any links. I am wondering if someone is using this to probe for sites that don’t delete comment spam, and then concentrate their efforts there.

Hopefully Google will kill comment spam. They (and hopefully other search engines will follow) have a new attribute to add to links so that they are not indexed for the purpose of calculating page rank. Hopefully this will remove much of the incentive for the comment spammer.

In my opinion, we need a well deployed cryptographic reputation system. Sadly OpenPGP and S/MIME have not been widely deployed. Hopefully some internet companies might someday bring easy to use digital signatures to the masses.

I just installed Gnome Blog. It is a great tool to allow me to blog directly from my desktop. Of course, it is in Debian.

I look forward to easier blogging.

Archzoom is now setup on boykin.acis.ufl.edu. You can use it to browse local GNU Arch respositories. It is also set up to serve tarballs, so you can get revisions even if you don’t have Arch installed.

Here is the local archzoom. You can also find a link to it on the right under “Resources”

As I recently mentioned, I have a strict policy for managing my home directory. A separate approach is to put all your files into a version control system.

For instance, here are instructions for putting your home directory in SVN. Subversion (SVN) is a version control system.

While the version control home directory idea has some nice aspects, as it is, it does not address one issue I have: cruft buildup. It would be interesting to look at the distribution of file access times, but my guess is that I use a file for a while, modifying, reading it, etc. but eventually I stop working with it and move on. My directory policy will eventually move that file out of my sight because I only see files I have edited in the last 1 year, the rest are kept in a directory for the year they were last editted.

I could certainly make a repository every year to put everything in and that would deal with that issue, however, there are other issues: binary files are not handled efficiently by most version control systems and generated files don’t need to be versioned. Though, perhaps I should give it more thought.

What is more interesting to me is integrating rsync into my setup more effectively. Hopefully I can improve my policy with specific comments on how to effectively use it with multiple machines using rsync.

Recently, with the help of Jesse Bridgewater, I developed a Directory Organization Policy. For years as a computer user I only had ad-hoc filing systems. This in turn made finding documents and backing up very difficult.

I have been using the new system for about 4 months now and it is working great for me. One of the things it specifies is to keep subdirectories in a top-level directory for the year that the last edits took place. This means for the new year, I had to do some business:

dirpol_skel.py (enter 2005)
chmod -R -w 2004/
rm current && ln -s 2005 current

Now, the old data is only readable, and the new directory is current. Since I am in the habit of putting things in the “current” directory, the shift should not be much of a problem.

In the next day or two (when I have consolidated all my files from a couple of machines) I will burn a CD (hopefully it fits) of all the data from 2004. Then, I know that data cannot change (since it is read-only from here on out). Finally, I use rsync to backup the current directory on a remote raid disk array. All the same, I should probably get in the habit of making periodic CDs of the current directory.

Finally, it was interesting to see that I am I am not the only one taking this general approach.