tugrik

I finally got around to figuring out how to have my Packetshaper selectively put the hurt on very specific parts of the link. It's always been able to bandwidth-cap based on URLs but I never quite figured out how. All it took was a little bit of RTFM and I've got it.

As an experiment, I had the 'shaper put the selective-squeeze on the following services: The Nice's UBBS forums, the Nice Picturebox and a handful of other bandwidth-heavy tidbits. The effect is pretty amazing, actually. Administrative access to the various boxen is now bearable, and both Lynx and Reality_fox noticed an immediate improvement in their services. All the other little sites are breathing a lot easier now.

The downside, of course, is the UBBS forums have slowed to a crawl. I'm counting the minutes before I get "J00 l0z3r! setz our forums free!1!1" user-hatemail. The picturebox isn't suffering too badly in comparison. As I'm watching the levels of rejected traffic it's suddenly very apparent just how high the hitrate is on that board. Terrence, the fellow who runs it, is making a good-faith effort in finding it a new home. After 24 hours of experimenting is over I'll let up the bandwidth caps a little so the forums won't be so painful. I think I may even have a way to be even more selective in what gets the pinch and I can start trimming away at the highest bandwidth forums only, so the low-traffic ones don't get penalized alongside.

I do have one longer-term plan forming now. Revar's made some new bandwidth analysis tools that are proving quite helpful. The 'grand idea' so far is to have a system that automatically rates sites we host based on their bandwidth usage. As sites go up in bandwidth and cross pre-determined thresholds, their owners will get email notes letting them know what's going on. If their bandwidth usage continues to increase they'll get bumped into the highest category which will be packet-shaped to keep them under control. This way everybody on the site will have equal opportunity at the bandwidth until they reach the same plateau... and once they reach it, their site performance gets ugly until they either find a way to reduce use or they find a more capable site to host at.

If this idea works out we'll be able to accept more entry-level candidates for new sites. As they grow the issue will take care of itself (at least from our end). Things will run smoothly until they get a warning and then traffic-shaping, letting them know it's time to move on. We'll see how this works out.