Tuesday, October 31, 2006

Floating-Point Addresses

Last week during the Wild & Crazy Ideas session at ASPLOS, I mentioned floating point addresses, and several people asked me about it later. I partially incorrectly attributed it to Jim Kajiya; the CS Tech Report on the Caltech Object Machine (COM) is by Dally and Kajiya (in that order).

The basic premise is that segmented architectures with fixed-size segments are always a pain because the segments are never the right size -- either the segment is too small, or if they're large, you can't have enough of them to be useful. In COM, the exponent of a floating-point number can be considered the object identifier, and the mantissa can be considered the offset within the object. Since the boundary between the exponent and the mantissa is flexible, you can have lots of small objects, or a few large objects, under system & compiler control.

It's been a while since I looked at it; I think objects are limited to a power of two in size, so bounds checks on them will prevent gross memory allocation violations, but not necessarily subtle off-by-one bounds errors. Other obvious questions I don't recall the answers to include how memory fragmentation is avoided.

I'm curious how this concept could be melded with modern compiler array privatization techniques. COM seems, offhand, to be very good for arrays but less useful for heterogeneous objects, and privatization could certainly help there.

The link above is to a Caltech technical report, but I believe they also got an ISCA paper out of it; I don't know if the two are the same. The TR talks about a machine simulation; as far as I know, no prototype was actually built, but I could easily be wrong there.

The Gates of Hell

I don't know a lot about sculpture, but in my humble opinion, Rodin's "The Gates of Hell" is the greatest sculpture in the history of the Universe. Stanford has dozens (hundreds?) of Rodin sculptures (including early maquettes of "The Burghers of Calais"), both inside and outside the Cantor Center for the Visual Arts, as well as scattered around campus. The museum is well worth a visit if you have the chance; it's open late on Thursdays and admission is free, so you really don't have an excuse not to go. The museum is medium-sized, and sort of eclectic, but contains a number of powerful pieces.

Wednesday, October 18, 2006

Analog Computing Papers

Three I've run across, but not fully digested yet:

title={The Complexity of Analog Computation},
author={Vergis, A. and Steiglitz, K. and Dickinson, B.},
journal={Mathematics \& Computers in Simulation},
comment = {Claims a digital computer can efficiently simulate an
analog one, and an analog one can't solve NP
problems if a digital one can't.}

title={{On the computation of Boolean functions by analog circuits ofbounded fan-in}},
author={Turan, G. and Vatan, F.},
journal={Foundations of Computer Science, 1994 Proceedings., 35th Annual Symposium on},
comment = {Haven't read yet...}

title={{On the effect of analog noise in discrete-time analog computations}},
author={Maass, W. and Orponen, P.},
journal={Neural Computation},
publisher={MIT Press Cambridge, MA, USA}

The Turan paper requires an IEEE Digital Library subscription to get, the others are available free on the web if you look.

Sunday, October 15, 2006

Have We Found Our Architect?

If we were to stay here for a long time, we would want to find an old house and renovate it -- the older, the better. In the last few weeks, the Daily Yomiuri's regular "Cultural Inroads" column has featured two foreign architects who live here and work on preserving traditional Japanese homes, updated for modern use.

This week's is about Karl Bengs, a German architect who lives in Tokamachi, Niigata-ken. His 180-year-old thatched-roof farmhouse now has pink walls, which I could probably do without, but looks fantastic inside. (This column is apparently not on the DY website yet.)

A few weeks ago, the column was about Geoffrey Moussas, an American and an alumnus of the MIT-Japan program who lives in a machiya in Kyoto that he reconstructed over a period of several years. Moussas is also a part-time lecturer at Kyoto University.

The number one thing I would want in remodeling an old house is the ability to keep a house dry -- I'm tired of finding kabi (mildew) on everything, including my shoes. I would want the house to be well-insulated on general principle, but I'm actually more or less okay with not every room in the house being heated to exactly the same temperature, as Americans are accustomed to. A decent kitchen (rare in Japan even in recent houses), a nice bath, and a Western-style toilet with a heated seat, and we're almost there. I want a small, well-lit Japanese-style room with lots of bookcases for an office (small enough that I'm forced to not clutter it up). The house has to have at least one tatami-mat room. And if it's on a lot big enough to have a small garden, I love Japanese-style gates and garden walls.

Anything else? Hmm, can I get it within bicycling distance of a nice shotengai (shopping district) with some local, historical color (like, say, Kawagoe, where we went to a fantastic festival today that I'll write about sometime), a nice beach, a go club, and the office, and I'm in paradise. Well, of course, except for missing my family and friends, most of whom live on a different continent...

Friday, October 13, 2006


Some researchers in China have just posted a paper on the arXiv, covering their implementation of a four-node quantum key distribution network, with a wave division multiplexing router in the middle.

Looks like interesting work, I'm looking forward to reading it in detail...

Thursday, October 12, 2006

Kashiwa Joban Art Line

November sees the arrival of the Kashiwa Joban Art Line, a month-long arts and music festival. Looks like there will be performance art, music, museum exhibits, and maybe some participatory things. The Japanese-only website doesn't seem to have an easy-to-read calendar, but all the info is presumably there...

Wednesday, October 11, 2006

Petabytes of Personal Data

A couple of days ago, I posted a note about Microsoft having 5.5 petabytes of crash dumps of the Vista release candidate, collected presumably from all around the world from members of their beta test program. If each crash dump is a gigabyte, that's 5.5 million individual crash dumps.

My first reaction, as an engineer, was to be impressed and a little envious. Even as we near a terabyte per spindle, building a multi-petabyte archive, collected over the Internet in half a year or so, and processing it is quite an accomplishment. It's an incredible engineering resource, and it must be fascinating to write tools that accelerate debugging by leaping from dump to dump, looking for data that will confirm or disprove a hypothesis about a particular problem. Certainly a problem related to a specific hardware configuration must stick out like a sore thumb.

My second thought, as a smug Linux user, was that it would take a really long time to get 5.5 million crashes, even if everybody in the world switched tomorrow.

Then this evening it occurred to me that Microsoft now has the memory contents of millions of people's PCs. I wonder what's in there? Bank account info? IM from a congressman? Crypto keys? It seems likely that Intel and Oracle have extensive beta test programs; perhaps part or all of a chip design or database product strategy?

You don't have to be a conspiracy theorist, or even loathe Bill Gates, to think that any one organization collecting the memory contents of millions of computers is a questionable idea. It has to be a tempting target for hackers, ambitious Justice Department folks, or even curious Microsoft employees.

I'm sure there are people out there who are members of the beta program. Were you made aware that your memory contents would be sent to Microsoft in the event of a crash, and were you warned not to use it for sensitive work? Was there and agreement you had to assent to? I wonder if this violates any EU privacy laws... Does anybody know the technical details of what is and is not included in a crash dump? (e.g., is the screen memory dumped?)

Monday, October 09, 2006

Now That's a Lot of Crash Dumps

According to the New York Times, Microsoft has collected more than 5.5 petabytes (5.5E15 or 6.2E15, depending on whether you're doing base ten or base two byte measures) of crash dumps from Vista Beta 2. That's at least 450,000 crash dumps, if I read the article right. Nah, it's gotta be a lot more crashes than that -- I really doubt each crash dump is ten gigabytes...must be five million or more individual crashes.

Analyzing that many crash dumps must be fascinating. I wonder if there are automated things you can do in the analysis when you have that many individual cases that make a qualitative difference in your ability to find an individual bug, or if the vast majority of the dumps are uselessly redundant. Certainly it seems likely that bugs exercised by particular hardware (or even generated in particular device drivers) should stick out like a flashing red light (or blue screen), and repeatability is the first step in fixing a bug.

Not a Fun Moment

North Korea has apparently tested a nuclear weapon. If I read the NY Times' description of the monitoring network right, the magnitude 4-ish tremor is near the lower bound of what's detectable, and that boundary is believed to be near 1 kiloton in yield.

It's theoretically possible to reduce the size of the tremor by building a test chamber designed to decouple the shock waves from the surrounding rock, but I haven't heard any suggestions that DPRK has done such a thing.

NHK is running an hour-long program about what is known about the test, and world reactions (especially the Japanese politicians). I wouldn't trade jobs with Prime Minister Abe, President Bush, or President Roh for anything right now.

I'm not prone to particularly bad dreams, but the news over the last few days about the buildup to the test must have gotten to me; I dreamt last night that I died in a nuclear blast. One of the worst dreams I can ever remember having.

Friday, October 06, 2006

Echizen Jellyfish in Wakasa Bay

Yow! (Yes, that's a diver in the background. The picture they put on the front page of today's Daily Yomiuri is actually a better picture.)

(If that picture has gone away, likely since Yomiuri's website is bad about archiving things, it's a picture of a diver in a cloud of meter-wide, brown Echizen jellyfish. It literally looks like the jellyfish scene from "Finding Nemo"; you can't see through the cloud. Apparently they were brought to the coast on the Japan Sea side by the recent typhoon.)

Thursday, October 05, 2006

A Really Big Piece of Pi

Akira Haraguchi has recited the first 100,000 decimal digits of pi from memory, according to today's Japan Times. Took him sixteen hours. This is impressive (or insane, depending on your point of view). I've heard of a ten-year-old who memorized the Koran, and people who have memorized the entire Bible (an astonishing feat), but those have structure. 100,000 random numbers, man. I can't even talk that long without going hoarse, and I like to talk.

Haraguchi hopes his feat will be listed in the Guinness Book of World Records. The current listing is 42,195 digits; last year Haraguchi went 83,431, but that one hasn't made the book yet.

Audio Bar Codes, 3D Displays

DoCoMo is advertising a new feature on some of its phones: audio bar codes. For quite a while now it has been possible to use your cell phone camera to snap a picture of a bar code and be transported to a web site with product information (including provenance of fish or produce, at least in theory; in practice it usually seems to be promotional material). Now they have a new way to send you small amounts of information, such as a URL: high-frequency sound. You're walking down the street assaulted by the usual battery of recorded touts begging you to come in and buy something, and now your keitai (cell phone) can get in on the fun, too, picking up a URL or two out of the cacophony. Of course, I'm sure they intend to see the equipment to broadcast such audio, too.

The ad I saw this in had a link to NTT's R&D website, but a quick glance there doesn't turn up anything about the audio bar codes. I did see a press release about a 3D lenticular display that tracks the position of the user to optimize the 3D-ness of the image, as well as one about a water-based fuel cell for cell phones being jointly developed with Aqua Fairy.

Volcanic Winter

A nice article from NASA's Science News mailing list, on recent analysis of the climate effects of Novarupta. Novarupta was the largest volcanic explosion of the 20th century, in the Aleutian Islands. Scientists are discovering that the effects of Arctic volcanoes are very different from the effects of tropical ones.

Tuesday, October 03, 2006

New Storage Research Mailing List

Garth Gibson has just announced the creation of a new storage research-related mailing list, intended for announcements of conferences, etc. This is something we've needed for a long time.

Monday, October 02, 2006

Mitsubishi Jet?

An article in today's Daily Yomiuri talks about an ongoing pre-development design project between METI and Mitsubishi Heavy Industries to design a 70-90 passenger regional jet, to be powered by Rolls Royce engines. ("Heavy" would seem to be an unfortunate word to have in the name of a plane manufacturer, but never mind.) A go/no-go decision is expected next year, and it will debut in 2012. Tens of millions of dollars have been spent already, but that's nothing compared to full-blown development and manufacture of a passenger jet. It's expected that sales would have to be 350 planes to break even, and 600 to make a decent profit.

The plane's selling point is expected to be that it will be 20% more fuel efficient than similar planes from other manufacturers, but the article doesn't say how that's to be achieved. It also says that MHI wants to work with trading houses to arrange favorable financing for export.

Japan hasn't produced a full airliner of its own since the YS-11, which ended production in 1973 after production of 182 planes. I'm looking forward to someday flying on a Honda Jet, though.

Wind Powerless

An article in today's Daily Yomiuri about a single large wind turbine in Gunma-ken is very frustrating. The headline says "Wind-power station generates losses", and goes on to talk about the money they've lost, but with such an odd collection of numbers that it's hard to figure out exactly what's what.

"The facility was projected to produce 327,200 kilowatts of electricity a year," the article says. Yup, that's what it says. Not "kilowatt-hours", "kilowatts". If we assume that they meant to say kilowatt-hours, that's a year-round average production of 37.352 kilowatts. That seems modest for such a large turbine (there's a photo, it must be at least eight or ten stories high, but it's hard to tell), but the article doesn't give its size or peak output rating. Worse, the turbine generated only 274,000 "kilowatts" in its best year, fiscal 2000, and last year generated only 192,000, less than sixty percent of the planned output.

The article also gives some economic figures -- saying that the Gunma government spent 76 million yen (about three quarters of a million dollars) and the New Energy and Industrial Energy Development Organization (NEDO) kicking in 57 million yen -- but it doesn't say what those costs covered. The losses are listed at five to seven and a half million yen (forty-five to sixty-grand, give or take) a year, but it doesn't say what price was being paid per kWh, what subsidies were in place, what the operating costs are, etc. Wind power in general still requires some subsidies to be economically competitive (until the price of coal and oil go up; it may already be competitive with nuclear, once government supports are factored in). Are the losses after those? Or do they not exist here? I'm not sure.

The article also talks about the "optimum wind velocity for power generation" being 13.5 meters/second. Of course, there is no such thing as a blanket statement; the power generated continues to rise as the wind speed goes up, until the turbine itself begins to be threatened by the winds. Certainly a wind that slow wouldn't be a problem. The article does say that average speeds at the facility have been no more than 3.6 m/sec. (and the power rises with the square of the wind speed, too).

"Its disappointing performance is blamed on a miscalculation of the winds expected in the area," it says. Uh-huh. Nobody knew that at Yoshiokamachi, between Mt. Haruna and the Tonegawa River, there wouldn't be enough wind. I'll bet a little digging would turn up interesting things in the site selection process. The article does seem to say that feasibility studies at Tsumakoimura, Showamura and Harunamachi showed negative results.

Ah, I did find a couple of sites (in Japanese) with some info on the project. It's a 300kw peak output turbine, built by Mitsubishi, and the blades automatically feather in winds above 24m/sec. I don't see at what speed the 300kw is output. One of the articles says this is enough power for 90 regular homes.

This one isn't it, but coming home on the shinkansen the other day, we saw a similar turbine deep in a valley somewhere in Fukushima-ken. It struck us as unlikely that such a location has strong enough winds regularly to make wind power attractive. I've also seen several near the port in Naoetsu, I believe, where it seems to make more sense. Japan as a whole is gradually catching on to wind power, and it's not uncommon to see new houses with small solar arrays on top, too. Maybe there's hope yet.

Steganographic Squid

This is wonderful: squid communicating via polarized light in a "hidden" channel. The commenter who said that they can do QKD may be thinking of a different kind of squid :-).