DISQUS

If you don't care about the specific protocol, why does XML have to be involved at all?

You could certainly do that in App Engine. Its image manipulation capabilities are limited, but it can certainly resize images.

I have a similar (half-finished but fully functional) web-service at http://imagin.notdot.net/ . It resizes images, and a lot more.

Absolutely agreed. I switched to using a DE razor a while ago, and find it so much better than cartridge razors it's difficult to even draw a comparison. I usually change blades once a week (I shave every other day), but of course that varies with personal preference. And as you point out, the blades are really cheap compared to 'modern' razors.

If you haven't tried it already, I'd also recommend getting a badger-hair brush and some proper shaving cream in a tub, for the complete 'old fashioned' shaving experience. It's much better than foam from a can. :)

Are you going to open-source it?

You haven't actually refuted his point, though - you've just pointed out that Obama used the system as it is.

Here (firefox2/linux), it works, but it doesn't appear over the content, it appears on a blank page of its own. Seems a bit pointless in that context.

But wouldn't you expect to pay more for 10 burgers than for 1? The amount you pay for your home connection does not cover the cost of that much dedicated bandwidth to backbone routers. Unless you're prepared to pay for a completely uncontended connection, one way or another you're paying by the traffic consume. Paying explicitly, like this, seems to me a much better scheme than everyone paying a flat rate, with those who don't use much subsidizing the small percentage of people who use way more.

I'm astounded by those claiming "250GB is very little" and "simply unacceptable". Though I don't currently live there, I grew up in New Zealand, where even now most connections are capped at 10GB or less (often much less), after which they ratelimit you to 64k or start charging you overage fees. 10GB is not that much for a month for a 'power user'. 250GB is _piles_.

Consider how many DVDs (30-60), or how many hours of 1200kbit h.264-compressed high quality video that is (20 days worth). Now explain to me how that's 'not very much'.

I'm sure you could build something on feedtree (http://www.feedtree.net/) with little to no effort .

Good point about the even distribution of the hashes. I'll have to remember that one. :)

Didn't you say that at least the subject will be a sequential identifier, though, and so not susceptible to that optimisation?

How many actual indexes will you need to efficiently support that set of queries, given your heirarchial index structure? Only 3?

I'd consider using an n-ary structure for your indexes instead of just a sorted list: Sort your index, then take every nth item (sized so the number of items returned just fits in a single disk block), and create a node from them. Recurse for each subdivision thus created. This will give you _much_ better locality of reference when you access the index, so for an infrequently used index, you won't have to read nearly as much into memory.