Cloud Computing Panel - July 29th 2008
Tonight’s Cloud Computing panel that I put together was held at the offices of Meetup Inc. (thanks to Jonah Keegan) and had great speakers and a great audience.


Our speakers were (from the left):
Mike Nolet - AppNexus
Jamal Mazhar - Kaavo
Hans Zaunere - New York PHP
Hank Williams - KloudShare (stealth - see Hank’s excellent blog)
Geir Magnusson - 10gen
Ronald Bradford - 42SQL
The panel started with an attempt at defining what cloud computing is. The term has become too popular and hence everyone started using it too loosely. So it now means different things to different people, but mainly, it’s the abstraction of services from hardware and scaling of data services properly so that services can scale to run on any number instances easily for different levels of demand.
But services do not move magically into the cloud and scale. Almost all applications have to be re-architected and rewritten. Engineers in the audience also noted that it requires a rather fundamental change in their thinking.
Different performance requirements almost always require a re-architecting of a service, the panel noted. For example, Ebay had been rewritten from scratch 4 times so far with completely new architectures everytime their performance requirements have changed. Here is a great article about ‘lessons on scalability from Ebay‘
In CC, ‘consistency’ almost always takes a backseat to ‘availability’. That is, it is much more important for a service to be available than for it to be consistent. For example, on Amazon’s SimpleDB, it takes around a minute for a DB write to settle across the system.
Data security is another big issue, and, for example, AppNexus is solving this for some of their customers by giving them dedicated hardware, so there wouldn’t be any ‘jumping out of the VM’ issues.
But eventually not every company with a website wants to have specialized system administrators on their staff 24×7. So, specialization of skills and economies of scale will lead to mass adoption of CC. Jamal noted that CC is now a $200M industry posed to grow significantly over the coming years.
Hank identified 5 different levels of creating a web service:
1) Rolling your own: buying a machine, colocating it, maintaining it, etc.
2) Traditional dedicated hosting.
3) Smart hosting: Amazong EC2, AppNexus, etc.
4) Tools layer: where tools help with the abstraction, for example SQS, S3, SimpleDB
5) Platform: Coghead, SalesForce.com, Google App Engine
Here, 10gen probably is between 4 and 5.
The panel predicted that levels 1 and 2 will eventually disappear or morph into levels 3 and up.
It was agreed that ’standards’ would help spur the demand side but the supply side cannot focus on standards yet since it is still ‘very very’ early in the industry and the innovation cycle, so it’s hard to predict how the industry will evolve and what will become the ‘de facto’ standard that might evolve into industry standard. Cloud computing is a rapidly evolving field that will bring us a lot of innovations and we are excited to watch it grow. Unype will definitely be using one of these solutions to scale.
Comments
One Response to “Cloud Computing Panel - July 29th 2008”Trackbacks
Check out what others are saying about this post...[...] or they could kiss their exits goodbye. This required buying/leasing servers, architecting, rearchitecting and then maintaining clusters of servers, replacing burnt hard drives, backing up, adding more [...]