I recently encountered some wwweirdness I’m hoping someone smart can explain to me.
So there’s a client computer A1, and a host C serving some web pages. If you ping or traceroute C from A1, the round trip time is consistently around 200ms. if you use a web browser on A to visit one of the pages that C hosts, then most of the time individual requests are served in a similar time.
But for around 25% of requests, it takes orders of magnitude longer to serve the file: around 20s. The behavior doesn’t seem related to the URL being fetched: all are usually fast, sometimes very slow. Logs on C always show responses being turned around rapidly once requests are received - the delay is never visible within the server side.
A1 is a machine in a domestic residence using a particular Net connection from a particular ISP. The problem is reproduced for another client machine A2 when it uses the same connection. C actually serves a variety of different domains, and the problem is reproduced across all of its websites, and when using its numeric IP address.
It appears that tens of other clients use C’s websites without encountering the issue. A1 is reported to occasionally see anomalous delays when browsing other websites, but nothing quite so remarkable - could be regular network unpredictabilities.
And finally, on overriding A1’s DNS to persuade it that awebsite is actually served by host B, and telling host B to reverse proxy all requests for pages under that domain through to C, then the issue disappears.
So it seems that A’s ISP and C just don’t get along with each other, but they do get along with the rest of the world. I can postulate some kind of blacklist on one side or the other, with a deliberate policy of throttling through introducing huge delays for a random sample of HTTP requests, but that sounds like quite a stretch.
(Especially since A and C are both perfectly reputable. Probably irrelevant concrete values for my abstract identifiers: the ISP for A is VirginMedia / Blueyonder. C is a virtual host I rent from Slicehost. B is another virt rented from Bytemark by a friend.)
Anyone have a halfway plausible explanation?
I have loved the Google Chrome browser from the moment I saw it. But my main machine at home runs Ubuntu, and the only OS Chrome is properly released on as yet is Windows. So I found myself bizarrely looking forward to needing to launch Windows to play a game or develop in MSVC, since I got the side thrill of launching something shiny when it was time to look webwards.
But lately I’ve been trying the unofficial daily builds of Chromium for Ubuntu and they’re working great. Even the Flash support is getting better (and it isn’t a big deal for me because my only usecase for Flash is video, and save it all up for a single viewing session after midnight due to the unusual bandwidth charging policy of my broadband provider.)
So for day-to-day browsing, Chromium is now my default choice. Yay!
(Ironically the main time I’m not using it is when I have to swap often between multiple Google accounts - e.g. personal and Modern Mixers - when using their services; in Firefox, there’s an add-on for that. The Fox’s extensibility - Greasemonkey, Firebug - remains unchallenged, and there’s no way I could do without it at work.)
LiveJournal appears not to have seen updates to any third-party feeds for around forty-eight hours. (It’s a known issue on their support page.) This impacts me, since I use LJ mainly as a feed reader - with very few exceptions, my “Friends” list is composed of syndicated feeds. So I haven’t seen updates from anyone I follow anywhere on the web for a couple of days.
Of course, I know that relying on any one service is a bad idea, because Sometimes Stuff Breaks. But LJ is all fluffy and open, and so my plan for an outage like this was simply to export my subscriptions as OPML and read the feeds elsewhere. But, nuh-uh. The OPML that LJ exports points back to LJ’s own broken endpoints for the syndicated feeds, not to the feeds themselves. This is so dumb! Time to start scripting and scraping, I guess…
When a major programming language conference comes to your home city, it seems churlish to ignore it. I have done precious little functional programming in the last fifteen years, and/but ICFP includes tutorials in the DEFUN development track, which ought to have met my hopes for abbreviated but intense introduction to new languages and concepts. So I attended two tutorials today - but they were kind of disappointing.
Apart from me, the first tutorial (on contemporary OCaml) seemed to mostly attract sympathy attendance from a few people who could quite happily have given the talk themselves. Since they were the majority, it was not unreasonable that the session was mostly derailed from its advertised purpose into a more academic ML family chat, but I still felt somewhat cheated - although perhaps not of very much, since what did survive of the original content seemed pretty poorly organized.
The afternoon session was substantially better, but I was still annoyed that Simon Thompson’s carefully prepared and well presented introduction to Erlang was too often interrupted by what seemed like culturally biased reactionary underinformed first impressions from some of the audience. The resulting delays prevented us from doing any of the planned coding exercises. So really I could just as well have read a book instead.
My overall verdict has to be that I wasted my time and money today. Sigh.
