» Cloud Computing: Dr. Kai-Fu Lee of Google
Main hall where
keynotes were held. I love the red slip covers on the chairs. They
were more comfortable than your standard hotel chair.
(click to
enlarge) |
The opening keynote at WWW2008 is Dr. Kai-Fu Lee of
Google.
Before the keynote, we were treated to a presentation that featured
dancers in blue Spiderman uniforms, a dancer in what I assume was
traditional dress, and a guy with a "Welcome to Beijing" banner
running through them all. Somehow, it seemed to fit perfectly even
though it was the first of it's kind at any tech conference I've been
too--especially one that's essentially academic.
We received a welcome speech from Dr. Yong
Shang who is the Vice-Minister of the Ministry of Science and
Technology. It basically said "thanks for coming, China's pushing
forward with Internet technology." No mention of the firewall. :-)
As an aside, the fact that I can find him and his ministry on Google
in English speaks louder about what he was saying than his actual
words. No doubt the Chinese government understands the power of the
Internet. That said, in terms of eGovernment, there was mostly
information there, not much in the way of services I could see.
The Internet connectivity has failed and we're not even 30 minutes
in. Hopefully it will come back up. I was planning on watching
Twitter for news of the Pennsylvania primary. The opening ceremony
has gone on for 40 minutes now. Finally we're ready for Lee's
keynote.
Cloud Computing:
Dr. Kai-Fu Lee of Google
(click to
enlarge) |
He starts out asking what people want. Many of his answers were
specifically about accessibility and it's control. There are four
key attributes of cloud computing
- Data stored in the cloud
- Software services are increasingly moving to the cloud and
accessed through the browser
- Based on standards and protocols
- Accessible from any device
Interesting that this is more or less the Google's core set of
beliefs. Companies often distinguish themselves from Google in departing
from these principles. The world has moved from hardware-centric to
software-centric to service-centric.
Six ideas driving cloud computing:
- User centric Data is stored in the could and follows you and
your devices. Data accessible anywhere and easily, safely shared with
others. He mentions several obvious examples of Google services that
meet this definition.
- Task centric People don't want to make spreadsheets or
write documents. Rather they want to plan a curriculum or
collaborate on a business plan. Right now, of course, all Google's
examples are simply documents or spreadsheet with collaboration
built in.
- Power Lots of computers in a cloud can do things you can't
do with a single PC. Google search is faster than desktop search
because there's lots of computers on the task. Cloud computing isn't
just about moving things off the desktop, but bring more data and
compute power to bear on the problem.
- Intelligent Intelligence comes from data mining of massive
data. "A ton of data is more valuable than an ounce of algorithm."
I'm not sure that says much. Machine translation is a good example
where feeding lots of good translation data into a learning
algorithm leads better translation of general text. Storage +
analytics = intelligence.
- Affordable Of course, this all uses a lot of computers and
that gets expensive. Google's strategy is to use cheap machines.
1000 CPU PC-Class machines cost about the same as on 64-way high end
machine and give 30x the performance (warning: data may be out of
date). The actual numbers at Google are even greater since Google
builds it's own hardware. Faulty hardware can be overcome with a
sophisticated software layer. This is the heart of engineering.
- Programmable How do you program 1000's of flaky servers?
Fault tolerant distributed disk storage, distributed shared memory,
and a new programming paradigm. Google uses GFS for file storage:
every piece of data is replicated three times. Anytime a server
holding on of the three chunks dies, the others notice and make
another copy. The shred memory architecture is Big Table.
The programming is done using MapReduce,
a way of creating parallel algorithms. Between Mar 2005 and Sept
2007 the number of processes using MapReduce went from around 72000
to over 2 million!
Cloud computing requires new skills. This is very true. We don't do
enough to teach these skills to students. We ought to be introducing
parallel computing in the cloud as the second programming
course--ensuring that the first emphasizes the building blocks for
the second. This probably means it's not in Java.
John Breslin has an excellent write-up of this speech as well.
Tags:
www2008
wgoogle
cloud+computing
web+services
