Web applications: wave of the future, or ripple?

[mdlbear]

Web 2.0 Conference celebrates Web app vision | InfoWorld | News | 2006-11-13 | By David L. Margulius

But one theme stood out: Web-based apps and services have become serious business, and everyone’s scrambling to provide platforms to deliver them.

“This is a fundamental architectural shift,” said Google CEO Eric Schmidt of the massive server farms necessitated by maturing Web development and delivery stacks. “The network is always going to be around; … the [local] disk will be optional.” He asserted that packaged applications can’t possibly compete against Web-based apps long-term because “the datacenter is running 7-by-24, it has to be better. It can’t break.”

I disagree completely with this idea. The network is not ubiquitous -- it would cost me an extra $200/month to allow everyone in my family unlimited net access through their cell phones. No way. Even with a T1 connection here at work, it takes three days to mirror a new architecture from the Debian repository.

And one of my coworkers carries Wikipedia around in an SD card on his Treo. All of it. A 2GB SD card is $40 at Fry's today.

Google's data center may be running 24*7, but LJ's certainly isn't, and neither is my net connection.

Opinions?

Comments

[trogula.livejournal.com]

As someone who designs and builds webservices (and the massive infrastructure that supports them) for exactly this purpose, it can be done, but it's not an easy task. It is only cost-effective for large scale stuff - you need very high speed switching and redundant routing to do this correctly, hosting redundant services in multiple data centers around the country (or world, depending on your target audience).

My smallest client's infrastructure services 9 million unique users a day, with a projected growth of close to 100 million by the end of next year (and we're well on target for that goal). This is massive stuff.

A T1 isn't ever going to cut it. Colo housing of services is the only way to go. You simply can't get the network redunancy and power requirements necessary for this at a home.

The power outage that LJ suffered simply shows how inadequate their operations staff is. This is a known, easily solved problem, and was totally avoidable. If I was managing lj, I would have immediately sent their head of ops packing for his/her incompetence. There is really no excuse.

In short, this will be done for services that can be used by a large enough audience of people, so that the operational costs can be justified. It is a scaling issue that is the barrier to entry here.

[mdlbear]

The problem is not that T1 is inadequate for the servers -- that's obvious. The problem is that for many of the suggested applications it's inadequate for the clients as well. Google Maps just barely works on my DSL line; Google Earth needs a local client.

And no matter how big, well-connected, and redundant your server farm is, you're still at the mercy of every router, DNS server and wireless access point between you and the client. Not to mention the client's poorly-maintained PC and spyware-ridden browser.

I'm not saying that web services can't work -- many of them work very well. Only that thick clients aren't going to go away. Ever. It may get harder to tell where the boundary is, though.