After Sunday’s brunch conversation, where we idly dreamt of a capabilities-based distributed-hash-table fault-tolerant filing system, Jeremy took a closer look at Tahoe, Zooko’s project that I’m continually plugging without completely understanding, and noticed that that is exactly what it is.
Last time I played with Tahoe I blundered into a dark alleyway full of unmet python dependencies and got mugged, but the latest version installed a lot more cleanly, and we now have a mini-grid going. I’m also re-reading RFC’s 3401 – 3405 on the Dynamic Delegation Discovery System. I can’t honestly tell you why it feels to me like these are all interlinked, because I’m still at that point where it’s all just pieces looking rather sad on the floor of my garage. I have good instincts, I think, but slow to rationalise, and I backtrack alot. Right now, for instance, my instincts are to put the DiSo people, the AllMyData people, the Zeroconf person, and several bored iPhone hackers in one room, lightly steam them in mild entheogens, and then just when they’re getting somewhere, divert a stampeding herd of straight-edge yet angry and dissolute teenage router-distribution developers into the building. And that can’t be right, so I’ve clearly got some more thinking to do.
I’m going to start dividing up these blog entries when I mention more than one thing, for the love of the metadata. But these are really throwaway items: we discussed how much power a home server draws, and I pulled figures from beyond my arse, but I’ve just run Kill-A-Watt on my old MacMini powerserver (the one running this website), and it does indeed pull down around 17W average. My arse speaks sooth. Of course, I’m not taking into account the cable modem, and the wifi router, so I guess I need to look into that.
In other news, one of the oldest bugs (but not the oldest!) that I watch in Firefox (@fontface support) is getting some new love. Soon, all the major browsers will be able to use downloadable fonts, and then what fresh desktop-publishing linotype hell will we be in?
August 13th, 2008 at 12:36 am
Re: @fontface. I have this open in a tab, no idea how I found it: http://opentype.info/demo/webfontdemo.html
I still have some .pfr files kicking around on my personal server from experimenting with this stuff ten years ago. I seem to recall it crashing Netscape 4 pretty hard on the SGI Indys at Uni. But then again, a stiff breeze caused Netscape 4 to crash…
August 13th, 2008 at 2:56 am
I love getting unexpected updates on really old bugs. A debian bug I filed in 2002 got resolved a couple of weeks ago and while I remember filing it, it definitely feels like a different person that did so. Like receiving a message from your younger self.
August 13th, 2008 at 3:00 am
“But then again, a stiff breeze caused Netscape 4 to crash.”
I imagine there are still thousands of still personal shell aliases out there still enabled that grep for the netscape pid in ps output and “kill -HUP” it.
August 13th, 2008 at 11:26 am
What kinds of things do you think DDDS might be useful for, even in the most handwavy terms? A way of discovering storage grids and other services? A way of mapping a file capability to some way to actually get the file?
I read the RFCs and my eyes glaze over before getting anywhere. “It is very important to note that it is impossible to read and understand a single document in that series without reading the related documents.” Indeed.
August 13th, 2008 at 3:16 pm
A sporadic point I’ve made at this here site is that a large amount of the data we have sloshing around these days is ‘commodity data’, which can be reimagined as key/value pairs, and the stuff that’s ‘personal data’ can be re-imagined as diffs. As Don Marti said in that linked thread: “Once you get to key/value instead of structured query, you can decentralize radically.”
August 13th, 2008 at 10:12 pm
Right, Nick, that’s a lot of what’s interesting about Tahoe, which uses that approach: identical data is (or can be) stored in identical “locations”. The key is derived from the content, so if you have the content you can find where its located (and if you need it back, you know how to retrieve it).
As it turns out, there are some privacy implications to “commodity data”, though, and not just the ones where the RIAA (or the US government) is suddenly very interested in what particular “commodity data” you share with their database of copyrighted works or terrorist reading matter.
For one thing, if I’m understanding the issue correctly, you can effectively do the equivalent of dictionary attacks on particular file variants. So, for instance, lots of people store root MySQL passwords in effectively the same format in a file called .my.cnf. You generate a huge load of .my.cnfs with a dictionary of commonly used passwords, and then find out who has that file (or is asking for it).
You can read: https://zooko.com/convergent_encryption_reconsidered.html for the gory details, and some ways to mitigate.