Archives week 41 of 2004
Oct. 11, 2004 - Oct. 17, 2004
While testing some memory-leak bug-fixes for PyOpenGL I discovered a memory leak in OpenGLContext. With a lot of tracking I found out that the leak was from Dispatcher.
Now, you may not care about that (I'm speaking to Bryan and Tim here), but Dispatcher is also used in Cinemon (which you will recall is currently ...!-->!-->
The hypothesis goes like this:
The two forms of memory, "lossless" and "lossy" have complementary function. Lossless memory tends to produce individuals who are more concrete (because they can recall random data associated with individual objects/situations), less creative and less abstract, while lossy memory tends to produce those who are better at synthesis, creativity and ...!-->
Met quite a few people. Shane brought... well... lots of people, including two ladies he'd met at a Second Cup. Very interesting both, one is a postmodern philosopher, but with a reasonable (read agreeing with me ;) ) perspective on the overall utility of the ideology. The other is a cultural anthropologist.
Both quite charming ...!-->!-->
Had another conversation tonight that involved someone having bought into the competitive model of society. This is the area of thought that is often described as "right wing", and includes Rand and Smith's "Guiding Hand".
It builds from the idea that the only way to accomplish anything is through force, with force including coercion. With ...!-->!-->
For each group (say 50 groups), for each field (6 fields), for each modem in each group (2500 down to 4 or 5), collect the min, max, mean, and total. Problem is, even summarising 1 field for a 500 modem group is taking a long time (i.e. a second or more).
I'm obviously just doing ...!-->!-->
Gather Bryan's been reconfiguring the demo instance to run against the larger CMTS (now scanning 2500 modems), so I don't know if there was any change in processing load between the limited and unlimited queries. Going to leave the changes in-place because it seems very likely that they will have a positive effect and no ...
Well, the memory leak (whatever it is) is down to about 50MB/day. That's still not going to work in a production environment, but it's slow enough that the demo can run for about 11 days (well, make it 10, it needs some ram available for actually interacting with users). It's also slow enough that the ...
I haven't had a chance to test this theory yet, but it makes so much sense...
When doing get bulk requests (which are done once for each modem on all non-intensive scans in Cinemon), the default behaviour for an agent is to encode 255 records for each requested field. That is, if you are requesting ...!-->!-->
Been working in Windows all day (good old reliable Windows ;) ). I'd forgotten how much more worn-in my working environment is here, everything set up for two-key access, everything defaulting to opening in the right directory, every tool winnowed from the pack over years of Darwinian selection, every device "just working", without few if ...
Did rather a lot of cleanup of the code today. Finished a number of "polish" elements (e.g. links upward in the hierarchy, retaining hierarchy names and comments through upload/download cycles, that kind of thing). Confirmed that email alerts do work when the email account is set up.
Last task of the day was sitting down ...!-->!-->