(Amazon)

Doug Kaye, building GigaVox Audio Lite. Lots of annoying things that you have to do to get a podcast sound professional, so GigaVox is an app where you drop in your MP3, and it’s converted and published. You can, if you want, put ads in and publish it.

Doug hardly did anything when he developed this – he used Amazon’s S3 Storage at the centre of it, and the EC2 service to do much of the processing. EC2 means that you can just use processors when you need it, and when you aren’t using it you don’t pay for it.

What if
– launching a new business on the web was simple?
– you only had to focus on the business
– You could manage growth more easily

What if you only had to compete on idea?
– That’s now how the world is right now, it’s a hassle to get things going.

John Hagel and John Seely Brown wrote a paper called From Push to Pull – Emerging Models for Mobilizing Resources.

What we are seeing online with the way that resources are used, these trends are happening in all sectors.

Technorati Tags: fowa07, fowalondon07

Push
– demand is anticipated
– top down design
– centralised design
– procedural
– tightly coupled
– resource centric
– restricted participation
– limited re-engineering

Pull
– demand is uncertain
– emergent design
– decentralised
– modular
– loosely coupled
– people centric
– unrestricted participation
– unlimited re-engineering

Drivers
– increasing uncertainty
– growing abundance
– intensifying competition
– growing power of customers
– greater focus on learning and improvisation

Not just web, also, say, education where students are pulling their own choice of modules together to make a course. University with highest satisfaction does this.

Push model
– design, deploy, execute and monitor, refine

Pull model
– find, connect, innovate, reflect

Find resources, connect those resources together, innovate on tope of those resources, then run the service and reflect on whether those resources do what you want and if they don’t you swap them out for something else.

Resource management in uncertain world
– acquire resources on demand
– release resources when not needed

In a push model, much of the work has nothing to do with your business idea, it has to do with infrastructure, anticipating growth, etc. Once you have a product you are working on, you are spending only 30% of your time working on it, the the rest of the time you are managing these ‘undifferentiated heavy lifting’.

If your datacentre gets wrecked, say in a weather disaster, your business is over.

How can you ever build an application that’s ready to scale form 100% to 1000%? Or provide 100% uptime? You need to be realistic.

Getting Real, book by 37 Signals.
Chapter 4 – Priorities – Scale later – it’s too hard to get right, so worry about the things that make your app better instead of thinking about difficult stuff.

Reliability of hard disks is not great. Failure rate is 8% in the first year. If you have 100s, that’s a problem.

Resources in pull mode
– fixed costs into variable
– scale up and down as your business does
– pay as you go
– leverage other’s core competencies
– focus on your idea

Amazon S3 (storage), EC2 (processors), SQS
– scalable, increase or decrease
– cost effective – low rate, pay-as-you-go
– reliable – runs on Amazon’s infrastructure
– simple
– compatible

Case studies
Smugmug photo sharing service, were spending a lot of money on building out their storage infrastructure, storing 120m photos by 2006. Moved to S3, saved $474k in first 7 months of operation.

Webmail.us, provide corporate webmail. use S3 to archive 10 terabytes of data, growing at a terabyte a month.

Second Life. Put instance of their download and put that on S3, and use it as content delivery. Stream 70gb per day for them. Offloads their own servers.

YouOS, Operating system inside browser, Java based. Sync your environment on start up and log out.

People build small tools around it, so that end users can use it, e.g. Jungle Disk, S3 Firefox Organiser.

Elastic Compute Cloud (EC2)
Virtual computing environment
scalable capacity, pay as you go

Use cases: Load testing, time or traffic-based scaling, simulation and analysis, rendering SaaS platform.

RenderRocket, on demand rendering, meant buying a lot of machines and planning ahead. But can fire up new instances as soon as they have a new customer.

Hanzo:Web, archiving and social web content.