A particularly odd bit of goofiness has hit the infosphere: cloud/utility computing mania. Nick Carr has written a book. IBM has announced, for the umpteenth time, a variation on utility computing, now called cloud computing. Somebody at Sun is claiming they'll get rid of all their data centers by 2015.
You know the flying car in your garage? The syllogism is:
- Google-style web-scale computing is really cheap
- Networks are cheap and getting cheaper fast
- Therefore we're going to use really cheap computing over really cheap networks Real Soon Now
Can you spot the fallacies?
Fallacy #1: Google is Magick The world's largest Internet advertising agency does have the cheapest compute cycles and storage (see my StorageMojo article Killing With Kindness: Death By Big Iron for a comparison of Yahoo and Google's computing costs). But they do nothing that the average enterprise data center couldn't do if active cluster storage were productized.
Google built their infrastructure because they couldn't buy it. They couldn't buy it because no one had built it. But all Google did was package up ideas that academics had been working on, sometimes for decades. Google even hired many of the researchers to build the production systems. Happy multi-millionaire academics!
Blame vendor marketing myopia for missing that opportunity. But their eyes are wide open now. If your enterprise wants cluster computes or storage you can buy it. From Dell.
Fallacy #2: Networks are cheap Or they will be Real Soon Now.
10 Mbit Ethernet from Intel, DEC and Xerox came out in 1983. A mere 25 years later we have 1000x Ethernet - 10 GigE - starting down the cost curve.
About the same time a first generation 5 MB Seagate disk cost $800. Today a 200,000x disk - 1 TB - costs 300 vastly cheaper dollars.
Also in 1983 the "hot box" - a VAX 11-780 - with a 5 MHz 32-bit processor and a honking 13.3 MByte/sec internal bus cost $150,000. Today a 64-bit, 3 GHz quad-core server - with specs too fabulous to compare - is $1300. Call it 1,000,000x.
Networks are the bottleneck, not the solution. Hey, Cisco! Get the lead out!
What's really going on? There are - currently - economies of scale, which Google is exploiting and MSN and Yahoo! aren't. So the latter two are going out of business.
But when you look at the cost of going across the network compared to the rest of infrastructure you realize that local - what we used to call distributed - computing is the only way to go.
Ergo, cloud computing will remain in the clouds and real computing will remain local. Where you can kick the hardware and savor fan hum and blue LEDs.
Sure, some low data rate apps - like searching - can move to the web. But if you want a lot of data and you want it now, keep your processor close and your data closer.
Comments welcome, as always.