Ha. Sometimes I forget about how lame the default clipboard is It’s amazing that almost all systems today still default to a single buffer clipboard. Mystifying.
Summary: Omniture‘s SiteCatalyst is really slick, but honestly is just too expensive for USC’s needs. It’s sort of surprising the how lacking the field of open source software is for log analysis. None of the most popular packages (Analog, AWstats, Webalizer) allow any clickstreaming or path analysis, much less any sort of mining/drill down.
After going through a bunch of packages, we acquired a license for one of my old favorites, Sawmill. It was the best for our needs, being able to handle large data sets without choking (I did testing on about 35GB of logs, less than a months worth), being useful straight out of the box and extremely powerful/flexible. Also it has the advantage of being dirt cheap. Some negatives: really clunky interface, sometimes you wish it would do better caching for drill downs you’re going to want, I haven’t quite figured out configuring the HTML output for all the drill down views, and a lot of options can be changed, but the entire DB needs to be rebuilt (not just added to) to apply the changes. On the 1-CPU 1.26GHz P3 I’m currently running stats on, that means usually, I don’t get an answer until a few days later.
Sawmill is multithreaded, which seems like it might be good for clustering, although right now it looks like the threads are per config, not for single large jobs…
Oh, other related stuff, I finally got around to installing and looking at ClickTracks today. OMFG, this thing rocks. In a way, it’s really obvious, but it is a really great idea. And a great implementation. What can I say? It seems to work. [60s viewlet tour]