web 2.0 expo notes

4

Click here to load reader

Upload: brian-moschel

Post on 25-Jun-2015

480 views

Category:

Technology


6 download

DESCRIPTION

Presented by Vlad Didenko and Don Albrecht at the April 2009 JS.Chi() meetup. A brief overview of the conference, the most interesting sessions we have seen and impressions from the exibition floor.

TRANSCRIPT

Page 1: Web 2.0 Expo Notes

• Past the ecosystem of frameworks.

• JavaScript quantum performance leap.

• Web workers (general purpose in-browser solution to backgrounding tasks) - enables responsive applications.

• Desktop integration - getting web applications out of the security box - gracefully!

• Microsoft rising - features, performance, compliance.

• New profiler in Firebug, FireUnit testing. Firebug 1.4a.

• YUI CSS Builder, BlueprintCSS for layouts now.

• Thunderhead is a layout system using JavaScript and Canvas for building Web applications (versus pages). Early Alpha.

• Atlas - Apple stack ported to the web

Web development tools

Page 2: Web 2.0 Expo Notes

• Autonomy - all decisions based on local information

• Asynchrony - always make progress

• Uncontrolled concurrency is evil

• Controlled parallelism - e.g. for loading

• Decompose - KISS, drop the fat

• Failure tolerant - failure is normal

• Local responsibility - do not burden your peers

• Symmetry - similar naming, makes configuration simpler of non-existant

Distributes services 1

Page 3: Web 2.0 Expo Notes

• Each service group was required to operate their own service. Amazon claims to have no operations for software services. Engineers motivated to do services monitor-able by getting 3:30 AM call. Usually a resource (page) build using 200-300 services.

• Amazon found 30% cost in value providing services, 70% in operations and infrastructure

• So to unload the 70% from engineers, which still will be supporting their services they did the EC2

• Scaling down is MORE important, than up. That is part of smart growth and being able to take advantage in consuming services improvements

• Failures are highly correlated. Failures come in batches

• Systems do not fail by stopping. They either crash or continue emitting garbage

Distributes services 2

Page 4: Web 2.0 Expo Notes

• When s... hits the fan, shed those services violating their consumer SLAs

• The infrastructure is virtualised and make it a service to developers.

• Before virtualization: 4-6 hours server provisioning. Engineers were NEVER releasing servers. They were hoarding their resources.

• The goal of EC2 was to provision AND RELEASE nodes in under a minute

• Service should break transparency. Engineers need to see failures.

• Engineers need to know location of servies to deploy resiliently

Distributes services 3