(this is page 32 of 121)





Latest

HTML5 won't cure cancer

O'Reilly Radar has a piece echoing a common sense of excitement for the collection of standards now traveling under the moniker HTML5. The piece argues that we can "finally" be excited about the web again due to all of the new possibilities that features like canvas, local storage, CSS3, and multi-threaded Javascript engines bring us.

While I too am excited for the promise of HTML5, it seems like the fervor is too much too soon at this point. First of all, it will take a few years to get consistent implementations across all of the browsers, especially when you consider a bear like CSS3 (or even what Microsoft might do). Second, while canvas and the video tags are great ways to finally kill off Flash, they aren't delivering new capabilities per se, since we've had Flash, Silverlight, and Java for years now.

And finally, while I love the idea of HTML5 as the default dev environment for the emerging smartphone ecosystem, it seems hard to believe, when even very prominent open web developers have moved over to the native dev kit for their work projects. Sandboxed though it may be, the iPhone SDK is still a lot more powerful than any approximation from Mobile Safari, and, unlike the desktop, there aren't spare CPU cycles and gigabytes of RAM to spend on a mushrooming web browser process.

Instead of hailing HTML5 as the cure-all, I wonder whether we might not be about to revisit the X-Windows system in a major way. That is, native applications that run virtualized processes on some server when they need to do compute and memory intesive stuff, but which receive a pixel-rich interface over the wire and send back clicks/gestures/keyboard entries. Google's Chrome OS is using an X-Windows replacement which is supposed to be faster and more efficient over the wire, so this may point the way forward.

Then we could have a gradual adoption of HTML5 features for the functionality that is mostly form-driven (looking up restaurants), native apps for games and other interactivity-rich or resource-constrained applications, and this new kind of native shim for running virtual blobs in the cloud for everything else.

While this approach seems like a bit of a throwback, it is worth noting that there are main grey-haired beards from the mainframe era running around (especially where I now work) talking about how the web with it's textual interface and request/response paradigm is really not that different from their VT3270 terminals. So maybe it's time we advanced from the 1950s into the glorious X-Windows 1970s!

Comments Permalink

Virtualization is the new black in the tech industry. What a handful of nerds reconstituted from the IBM 360 days at VMware, everyone wants to do now, whether it is Xen at EC2 (and all of their competitors), or KVM extensions or even Wine. Today, Eddie sent me a link to a company that looks a lot like VMware in the cloud— Xenocode— except that it's 2009 and not 1999, and people are still having pissing matches over whether the desktop is truly over and AJAX goodness is the only path forward.

Xenocode seems to be able to do per-application virtualization, wrapping your exe into a file which extracts itself into a sort of chroot for Windows. With it developers can run fully isolated Windows apps from their disks or from USB sticks, while the apps themselves think that they are running on a full host.

That's a neat trick, but the company will do you one better, as they can also stream these apps-in-a-file to a browser plug-in effectively giving you desktop-application-from-the-cloud for free.

Now I haven't used Xenocode enough to know whether it actually works as advertised, but the one thing that is clear is that we're heading into this totally virtualized Zen Zone where compute cycles will just exist in our personal fabric ready to run our various VMs for us. And whether it is an NES image on an iPhone, or a Win32 one running WoW split between that same iPhone and an EC2 instance in the cloud will not matter in a few years time when we've got enough LTE (4G wireless) to make up for our crappy networks, and seamless compute cycles for rent from many providers.

One thing is for sure: compute architectures are a-changing, probably more so than any time since x86 won. And it will be up to us to figure out what to do with them.

Comments Permalink

More holes in the clouds

Renting compute cycles and storage from the cloud looks so good these days, it's got to be bad. Or at least worse than all of its proponents argue these days. Yesterday I came across an interesting observation in "The Hosting Provider Time Machine: Paying 2006 prices in 2009." The author uses the Wayback machine to discover that the "slice size," effectively how much CPU, RAM, and storage you get for your monthly bill has remained relatively constant since 2006 despite huge advances in Moore's Law and whatever we call the equivalent thing in storage.

Now when a vendor is going to hide price increases in front of the ruthless cost plummeting machines that are Intel, Hitachi, and a host of Asian ODMs, they better bring with that effective price hike a tremendous amount of value. Apple, for instance, keeps margins high by always dropping prices slightly more slowly than the component costs (which has become obvious now that they are on a standard Intel platform). But in return you get $600MM annually spent on software R&D, industrial design, and great integration.

Even the best of the VPS providers, Amazon with EC2 and S3, hasn't aggressively dropped prices along with component costs. And when they do, as with the new reserved instance pre-pays, it seems to be more for marketing reasons than for passing on cost savings.

If this dynamic doesn't change, the cloud providers face an even bigger problem in their lack of network effects. There are goods and services where the per-unit price can remain relatively stable in the face of dropping costs because additional value comes from the growth of the underlying network. eBay is the classic example: fees remained relatively constant (even went up) despite a scaling infrastructure because the network effects made each auction more valuable. To take another example from outside technology: JetBlue has kept its ticket prices relatively stable despite bettering economics on the fleet (financing from Airbus, deals with the airports, etc.) because the value of the ever-expanding route network offsets their having to pass on cost savings to remain competitive.

As far as I can see, none of the cloud providers have yet figured out how to exploit these network effects. Even Amazon offering free bandwidth from/to S3 is relatively minor compared to say, transaction volume in a marketplace.

And unless they do, keeping prices artificially above component costs is just going to open a huge hole for large vendors who 1. must "be" in this business (Microsoft, IBM, HP), 2. can run break-even even at huge scales for a very long time, and 3. already have the commoditizing dynamic they need in the form of the x86 instance as the unit of computation and the GB/month as the unit of storage.

If I were playing in this space, this is one trend I'd be paying very close attention to.

Comments Permalink

Interdisplinary futurecasting

Just got around to listening to a recording of Charles Stross and Paul Krugman doing a joint session on the future at this year's World Science Fiction convention. It's an absolute blast to listen to.

Though I am only a part-time Krugman fan, I do find him quite insightful, and when combined with Stross's great imagination, you get something that is part economics, part sociology, and part future prognosticating.

My favorite part: hearing Stross telling Krugman about the novel he is currently working on which is all about the future of desktop 3-D replicators. Also entertaining: listening to the two of them talk about the shift from wealth creation to wealth concentration over the last twenty-five years, and Krugman explaining to Stross that the excessive compensation in the finance sector can be traced back to the start of the televised Monday Night Football era (listen to the Q&A to hear why).

Universities and conferences should put on more of the interdisciplinary sessions— they are a blast.

Comments Permalink

What we do with our time

In an absolute tour de force on the art of visual display of quantitative information, the Times has a great interactive chart on how Americans spend their time. Their headline is:

Sleeping, eating, working, and watching television make up about two-thirds of the average day

but that barely scratches the surface of what you can glean from the data, especially after you start playing with their demographic and psychographic filtering options.

The graph also proves half of Shirky's "cognitive surplus" hypothesis regarding the amount of television time that could be cannibalized for more productive "lean-forward" Internet-enabled activities (such as editing Wikipedia pages). Unfortunately what the it also seems to show is that "time at the computer" is still dwarfed by television for most of America. In fact it is quite amazing to see how many other activities: family time, socializing, traveling, are also miniscule from a time perspective compared to television.

Seeing this interactive graph, it is hard for me to think that television won't be remembered as the great time sucking parasite of the 20th century.

(Incidentally, I think that these interactive Flash graphs and charts are the best "content" the New York Times has going for it these days. I'd dump half of the written sections for interactive charts if I were the editors there and wanted to survive the Great Newspaper Meltdown of 2009)

Comments Permalink