Slashdot Mirror

Keep in mind that Yahoo can deliver unlimited bandwidth much cheaper than a hosting company can. You have to keep in mind that Yahoo has an expansive network and they are doing settlement-free peering with all of the Tier 1 ISP's, as well as anyone else who happens to be hooked up to a common peering point. Hell, even at our regional hosting center we're connected to a peering point and we peer directly with Yahoo, bypassing the Internet.

The point is that all that bandwidth doesn't cost Yahoo nearly as much as a traditional hosting provider would have to pay for it.

People, this is just clever spin. The entire industry is moving towards putting applications back behind the glass (where they usually belong). Sun's got some kickass virtualization tools, and the network is now ubiquitous. All this announcement means is that they're going to cut costs by outsourcing their data centers. Big deal. There will still be data centers, servers, system administrators ... but they won't be at Sun. Lots of companies outsource their data center operation. I oversee network operations for a hosting company in New York state, and I can tell you with certainty that demand for data centers is not slowing down. The applications have to live somewhere. Can you save money by having someone else run it for you? In many cases it makes economic sense, and Sun is going to try it.

Clever spin. See how they made everyone turn their heads and take a curious interest? How much better was that than announcing "by 2015 we're going to fire all our IT staff and farm out the data center ops to some third party" ??

Has business picked up so much that companies are hiring dyslexic people as web designers?

I work in the hosting business and I can tell you, there's definitely something happening. Back in 2001 when the bubble burst, we lost 40% of our customers in a six month period, and then business was downright anemic for a long time. Gradually we've been building up new business by signing more customers whose revenue didn't depend on "dot com" type business. This year, though, it seems that there's another ramp-up happening. Our cabinets are filling up again -- not at the irrationally exhuberant rate that they did in 1999-2000, but at a more careful pace. And customers are being careful with their spending this time; they're only buying the services they need. Lots of colocation this time around instead of more expensive managed hosting, for example. But it's definitely happening.

Let's hope that this time around, the "Internet economy" can get firmly on its feet.

Applications moving behind the glass. Any application accessible from any location, without having to load it on "your" computer first. Basically, it's a return to mainframe-like computing, but without the green screens.

Well-designed hosting environments can make this happen. Portable API's such as those available in Unix/Linux and in Java help make it happen, and help make the apps relocatable. Truly transparent network filesystems like NFS allow for application and server load balancing. Transparent graphics systems like X11 help make the apps truly independent of the display they're viewed on -- applications moving to the Web is a big piece, too.

This was the original vision of "network computing" and it's still a good idea -- it's still being worked on and there are places where it's being deployed. The reason why the original McNealy/Ellison vision of network computing failed is because they required everyone to move exclusively to pure Java applications. In reality, most environments can't make that big of a move that quickly.

So what we're seeing is a gradual shift of applications off the desktop and back into the data center. For the time being, most users are still using a fat PC to access them, but IT organizations will wake up one morning and suddenly realize that everything has moved behind the glass and they really are in a utility computing environment. If they've done it right, they will then be able to move applications and storage resources around the data center without an impact on the users. This is the promise of utility computing and it's a good idea.

And for organizations that don't want the expense of running their own data center, they can enlist the services of a hosting company that specializes in this type of thing -- IT keeps control of its applications, while someone else keeps the air conditioners, UPS's, and routers running.

There is no shortage of "small generator" capacity. The problem is with the local power grids.

We have three megawatts of power generation capacity, but we don't need all of it (our power needs are less than 1.5 megawatts; two generators are present for N+1 reliability). So we wanted to sell power back to the grid, and the power company wanted to buy it. But it couldn't happen, because the local grid in this area is not capable of accepting a backfeed. This is the problem in most places. There are probably tens of thousands of places with local backup generators that would be capable of supplying power to the grid, but until the local grid is upgraded to handle backfeeds, it simply can't happen.

What does happen, though, is that on days of very high demand, the utility will provide cash incentives to companies with their own generators, to voluntarily get off the grid and run on their own power. We did this for a couple of years. But ever since "deregulation" put utility prices through the roof, it's actually been cheaper to just run the generators 24/7. Diesel fuel is less expensive than the utility, which IMHO is proof that deregulation doesn't work... at least not when the White House is inhabited by someone who cares more about the welfare of energy companies than about the citizens.

This is kind of weird. I work at a mid size hosting center and ISP and since we normally run on generator power, we didn't know that there was a widespread power outage. Things started popping up on our monitoring system -- and they all seemed completely unrelated. Of course, it turned out that all the things going red were customers with T1 lines and such, that were in buildings losing power. :)

Ok, I read through the article and came away with absolutely no information. He says some things we already know: data centers are expensive, IT people are overworked, and the rest of an organization only notices the technology folks when something breaks. So, what should we do about it?

Well, here's where you expect an innovator like Andreesen to come up with a brilliant idea that's going to begin the next IT paradigm shift, but all he says is that we need to find some revolutionary way to automate our own stuff -- basically, to automate the act of automating things. And how? Well, he doesn't really know. He makes some vague reference to sending out automatic updates to hundreds of servers at a time, and that's it.

Real bright there, Marc. Automatic patches and updates. As if that's the answer. In the real world, you don't have a huge farm of servers that all run the same patchlevel of the same operating system. I've got a few hundred boxen behind the glass, for example, that are a mix of Linux, Solaris, FreeBSD, Windows 2000, and Windows NT. And I'd guess that at least 50 percent of them would experience some sort of problem if we were to just push updates out to them unattended -- different applications require different patchlevels and break on others.

Let's not forget the fact that there's more than just servers. There's infrastructure such as routers, firewalls, and switches. And of course there is the dreaded desktop, which is probably the source of 90+ percent of IT headaches. Until the IT world wakes up and gets the hell off local desktops, the maintenance nightmare will continue. Seen what Microsoft is doing lately? Their vision of the future is one in which applications are loaded through a browser and executed in a local .NET environment. It's basically the same as Java applets, but they call it "Smart Clients" to give you the impression that it's something they invented. Sounds a lot like Network Computing to me -- which simply means that Network Computing is a good idea after all! And now that Microsoft has "invented" it, the idiots who make up most of the world may finally start to adopt the idea. Make the desktop a stateless device like it was 20 years ago when we all had dumb terminals on our desks, and IT overhead will drop like a rock.

The other trend you're going to start to see is outsourcing. People are realizing that it's expensive to build and run a data center. Fortunately, you don't have to. All you have to do is run your servers at a hosting center that knows how to do outsourced IT (as opposed to just hosting web sites, like the first generation of centers like Exodus did).

There are ways of streamlining IT after all. Unfortunately, Marc Andreesen didn't touch on any of them. I give this article a "C minus."