Archive: June 21, 2003

<<< June 18, 2003

Home

June 22, 2003 >>>


JPEG2000

Saturday,  06/21/03  12:12 PM

I'm back!  I know I've been neglecting you; more travel, and much work to do...

I've been deep in the intricacies of JPEG2000 codecs, rather arcane but for Aperio rather critical.  It turns out that - like virtually all specs - the JPEG2000 spec is not 100% specified, there is enough wiggle room for different implementations to be, well, different; and the differences result in incompatibilities as well as perceptible differences in image quality.  Aperio's ScanScope has superior image quality to any other slide imaging solution, and so naturally we are anxious to have the best possible compression technology.

As well, different implementations have different performance characteristics.  JPEG2000 is highly asymmetric, which is a good thing; this means compression is much more CPU-intensive and takes much longer than decompression.  That said, we still want compression to be as fast as possible.  We have big images (like 10GB!) and they take minutes rather than seconds to compress, and the faster the better.  Some codecs have really good image quality but are so slow as to be unusable in actual practice.

The best solution for compression performance is hardware acceleration, and here again there are differences; some codecs are optimized for single-threaded performance, while others are designed explicitly for the parallel processing possible with hardware.  I want a codec which has terrific image quality, is compatible with all other implementations, runs fast in a software-only implementation, and is optimized for implementation in hardware.  Is that too much to ask?

 

Saturday,  06/21/03  12:30 PM

GNXP links an interesting paper from 1999: Sex Differences in the Distribution of Mental Ability.  Good reading if you're into this kind of stuff.  Makes the case that for humans, as for many other animals, it is to males' advantage to exhibit more variation than females.  Essentially males play a higher risk game for higher stakes; a male can have many more offspring than a female, but can also be more easily "shut out" of reproducing altogether.  [ via Razib ]

Males have a Y chromosome, of course, as well as an X, while women have only Xs.  This means Xs can "recombine" easily to form new mutations, but Ys cannot, as there is nothing to recombine with...  However we just noted increased variation is much to male's evolutionary advantage.  So what to do?  Well, how about recombining with itself?  This is apparently exactly what happens!

An interesting article by Sahar Akhtar in Salon argues that the Apple Music Store will hurt musical innovation, because it enables consumers to buy single songs.  In the album format artists can mix "commercial" works with more experimental pieces. 

This is a well-defended point of view, but I think overall the store will increase innovation, for several reasons.  First, while the store sells individual works, albums are also available and at a substantial discount (usually around 50%).  As long as at least half of the music on an album is interesting to a consumer, they will buy the album, and if less than half is interesting the album should never have existed anyway...  single hit albums have been the bane of the music industry for a long time.  Second, artists will release single works whenever they want, instead of batching them into a "commercial" album.  As people get to know an artist anything they release will be of interest, especially since people can preview the track without buying it.  Third, many more artists and styles will become available, instead of being funneled through the narrow commercial filter of a label.  Individual listeners will decide what they want to hear, instead of labels and radio stations paid by the labels.

{ I wonder what Ottmar Liebert would think...  actually I'm almost sure he would agree. }

The other day I discussed Ottmar's musings about a handheld PAR (personal audio recorder).  Well, now we have a head-mounted PVR (personal video recorder); it records the last 30 seconds of video and audio you've encountered.  Why not?

Oddpost is now an RSS reader.  A very cool web-browser-based email client which is now an aggregator as well.  Check it out!

Tim Oren tells us more than we want to know about VC disclosures, IRRs, and the J-curve.  I don't understand it all, but what I do understand is that there is more to venture returns than a simple percentage :)

Andrew Anker on the New Platforms: "If Solaris is still a viable development platform, it's certainly not obvious from here."  Or from "my here", either, i.e. CTO of a software startup.

Honda has introduced a new crash-detection system for one of their vehicles.  Awesome.  The technology for caravans is converging!

Speaking of cars, Philip Greenspun muses on the Chinese Car.  "Within 10 to 20 years the Chinese will sell a car that is similar to today's rental car: 4 doors, 4 seats, air conditioner, radio, new but not fancy.  It will cost between $2000 and $3000."  Exactly.  But will it have caravan technology?

Want a new pet?  How about a genetically engineered zebrafish that glows in the dark?  I am not making this up...

 

Windows Paging

Saturday,  06/21/03  09:19 PM

Does anyone understand why Windows paging is  s o   s l o w ?

Paging is the biggest reason why Windows is so much slower than other operating systems like Linux.  But if that is so, surely I'm not the only one to notice?  And if I'm not, why hasn't someone in Redmond reverse-engineered Linux' paging and adapted it for Windows?  The problem is the same regardless of the system's APIs and GUIs, so a better paging scheme should just, er, be better.

I just spent fifteen minutes waiting for Windows to page in a large program.  I had been running this program (an image viewer) and then loaded another large program (an image compressor) which paged the first program out.  After the new program was done, I activated the old one, and its working set of 512MB took 15 minutes to page back in.  Meanwhile loading that program from scratch and filling its 512MB cache takes less than one minute.  This just does not make sense...

On a Windows box you "fall off a cliff" as soon as the working set exceeds the amount of physical memory in the machine.  (If you don't believe this, you obviously don't manipulate large images.)  And once you're paging, Windows seems to take forever to pull stuff back.  There just isn't any obvious reason why it should be so bad at this.  Meanwhile with Linux you can routinely run with a working set twice the size of the physical memory, or more...  and there is no falling off a cliff, performance just gradually degrades as paging takes place.  This is one reason why Linux will run well on a box with less memory, it just doesn't have to have as much physical memory to support a certain size working set.  (There are lots of other reasons, too, like the fact that the Linux resident kernel is about ten times smaller than Windows.)

Macs used to have slow paging, too, in the old days before OSX.  Slow but not as slow as Windows.  But now with OSX Macs have a Unix kernel, and they have fast paging.  This is another reason among many why Photoshop runs better on Macs, you just don't have to have enough memory to keep everything loaded.

You would think that this is not a subtle fact.  You would think that with all the really smart talent at Microsoft, someone would deal with this.  A faster paging scheme would immediately make every Windows system everywhere much faster.  (I wonder if there is an opportunity for a third-party accelerator, which patched itself into Windows and replaced the paging algorithms?)

So, I'm really curious, does anyone understand this?

 
 

Return to the archive.