Wednesday, March 02, 2011

iPad II event is rolling out

iPad II event is rolling out right now. It seems there's nothing earth-shattering like a retina display this time around. But it is faster, lighter, and slimmer, all very welcome changes. And of course it now has cameras.
Apple will probably put up a video stream of the event tomorrow.
Update: yep, here it is.
And here is an intro video to the iPad 2.

In short I'd say that if one really wants the video chat cameras or the speed bump, the iPad 2 is nice. But on the other hand, iPad 1 is now an even better deal at only $399...

So now the question is if the third generation with a high-rez display is coming in fall like some rumors say, or only in a year? I hope for the fall. While the present iPad display is very good, the iPhone 4 Retina display with over 300 pixels per inch is simple awesome, especially for reading, text just leaps out at ya.

Another thing: this seems to be true of hand-held devices in general, though I have no idea why: they seem to download big files much slower and with much less certainty than desktop computers. It's very odd. I tried to download the Apple Keynote (850MB) to my iPad... after nearly an hour it failed, having downloaded and then lost 2/3rds of the file. I've now downloaded it to iTunes on my desktop machine (for transfer to my iPad), and it took less than five minutes, hitch-less.
It's weird: I can understand why the smaller processors can have trouble using big files (playing large videos, running big apps), but why should they have trouble downloading the bigger files? (Update, it seems to have to do with the writing speed of flash memory, see comments.)

Who misses the eighties?!


Kentg said...

Check out:

Bert said...

Re why the slowness of downloads to hand-held devices, our resident expert Bert wrote to me:

It's a complex and delicate subject. Although I have been out of the electronics arena for a while now, I haven't heard of any game-changing breakthrough on that front either, so the problem is likely to be the same as it has always been with Flash memory (and thus all solid-state storage): reading may be lightning-fast compared to magnetic media, but writing is a non-deterministic, error-prone process.

It would be difficult to explain why without looking into the actual workings of a Flash memory cell (bit), and we don't really want to go there. Let's just say that a bit is stored inside Flash memory by altering the static electrical charge on a microscopic piece of metal (referred to as a floating gate) buried inside insulating material. When there's enough of a charge on the floating gate, the bit reads as "0", otherwise it reads as "1".

You might rightfully ask how can the charge be altered on a floating gate if it is buried in insulating material? Well, therein lies the crux of the matter: the writing process (i.e. the changing of the state of a bit) relies on marginal physical properties of the dielectric (insulator). To further complicated things, erasing the bit (returning it to the "1" state) employs a radically different method than what is used to program a bit to "0". Anyway, that's the realm of physicists, and I gladly will let them explain the subtleties of avalanche breakdown, charge tunneling, and other similar phenomenon.

There's a lot to be said here, but it all leads to the conclusion that Flash memory is a cumbersome device to use. You cannot erase a single bit, erasing has to be performed on blocks of bits at a time (more-or-less large chunks of the chip, usually referred to as sectors). But bits can also be over-erased, leading to catastrophic failure, so you have to program all bits in a sector to the same state before you erase that sector, which can be time-consuming.

Changing a single bit from "0" to "1" thus requires an entire sector to be programmed to all "0"s, then erased back to all "1"s, and then reprogrammed to the desired contents. On the other hand, changing a bit from "1" to "0" requires the programming of only that bit. You may see how this asymmetry can make it difficult to predict the time it will take to alter the information stored on a flash device.


Bert said...

And here comes the real bummer: programming a single bit to "0" is an iterative process, you have to keep trying until you succeed (or give up). The floating gates mentioned above are insulated, but this insulation is only a few atoms thick. Programming a bit requires the tunneling of charges through this extremely thin dielectric barrier. Hit it too hard, and you'll punch a permanent hole in the insulation, so you have to be gentle about it. So gentle in fact that any variation in the insulator properties will lead to significant variations in terms of the time it takes to tunnel enough charges to alter the state of the bit, hence the need for the iterations mentioned above.

The behavior and requirements of Flash memory are in fact so different from those of magnetic media that it has led to the development of highly specialized file systems dedicated to solid-state storage devices. One of the primary goals in such file systems and associated OS software is obviously to try to smooth out the performance issues. From my own experience with early Flash devices, I can tell you that brute-force management of Flash storage leads to write performance no better than that of a floppy disk (yes, it is that bad). That is quite acceptable for infrequently updated storage (firmware, etc.), but certainly won't meet the expectations of iFad users.

Now, the most obvious way to increase write performance is to rely on cache memory (e.g. RAM, which can be written to at full speed) and background automation for the actual writing to Flash. Such strategies will yield excellent results for small files, but the cache memory rapidly becomes insufficient when handling larger files. In fact, writing huge files can be so excruciatingly slooow that I wouldn't put it above some designers to reduce safety margins to a minimum (thus reducing the number of write iterations required, see above) when handling multi-media files, because such files are generally considered to be more fault-tolerant. After all, what's a little clicking in your music, huh?

And we haven't even addressed the issue of device aging (repeatedly punching holes through insulation does take a toll over time, no matter how gentle you are about it). Nor failure management. Etc.

Bottom line: don't believe all you hear about solid-state storage, it's not quite the panacea you heard it was.

Hope this mumbo-jumbo helps a bit,


ttl said...

There is a second, slightly different iPad 2 intro video now out.

eolake said...

Pretty funny: flat, generic satire poking fun at flat, generic promotion. :-)