pre-talk

– next month: hypervisor showcase

— ms hyper-v

— vmware

— citrix xensource

– upcoming

— 3/3 java sig: running java apps on ec2

— 6/3 security sig/cloud services

— 40% off etech w/ code: et09ffd

— oscon in san jose

— interop

— salesforce talk

— cloudcamp?

— new version of app engine released last night

 

right scale

– michael cramdell, ceo of rightscale

— cloud deployment platform

— 10k users

— 300k servers

— features

—- automated management: reduced admin overhead

—- cloud neutrality

—- transparency

— sits btwn amazon, gogrid, rackspace, etc and cloud apps

— customers: animoto, sling, woot

– history

— pre-public w/ amazon cto 

— taught grad class w/ ror apps launched on ec2, but students would leave servers up.  so rightscale was built to automate instance spin up/down

— motivation

— forking a server

— pay as you go. foundations google and yahoo have built(!)

— just launch a server

— cost analysis

— 1 server for 1 yr on ec2 = $876

— 1 programmer for 1 day = $400- $1000

— throw servers not humans at problems

— classic dilemmas

— over building hardware = wasted money

— peak vs. avg utilization

— new mindset

— plan for failure

— focus on automation

— servers are disposable

— great for business

— 45 day float on bill

— no need to buy hardware

— no techcrunch

— ideal for spike-y usage

— barriers to customers

— fear of lock-in

— security

— loss of control

— overhead of moving

— customer needs

— fast onboarding

— expertise

— portability

— layers

— saas: most lock-in

— paas

— rightscale

 

– overview of deployment

— 4 x app servers

— launched dynamically

— httperf used to simulate load

— rule based scaling

— cloud computing is built on the shoulders of virtualization

— amazon uses syslog to create persistent logs from virtual servers

— cost = time to change x live images x time to retrieve, launch, change, x stored images

— server templates are base images w/ dynamic personalities, app server, web server, grid comput, etc.

— centos base images 

— on launch, download a list of programs that define personality

— clusters are based on templates, not hardware

— backend processing

— animoto has a web front end that farms off video rendering to grid backend

— failover

— for DBs,  promote slave and launch new slave

— amazon’s ebs volumes?

— mmulti-zone deployment

— all pop stacks supported

— animoto scaled from 50 to 3500 servers in 3.5 days

— clouds surpase traditional hosting

— multi server sizwes

– reasign ips

— multi failure zones

— multi geo

— dosk volumes

— layered arch

— cloud neutrality

— automation engine

— temopaltes

— apis

— web ui

— agile deployment

— partner oriented company

— ecosystem:

— rightscale under – customers, software vendors, cloud providers

— email michael@rightscale.com for a free tshirt

– qa

— do we need an account w/ amazon?  yes. 

— whats are the common features in the base image? 

— ec2 -> launch a server

— bigtable, sdb -> proprietary = lock-in

— scaling relational dbs?

— sharding dbs

— not yet fully solved

— most customers use traditional relational DBs for legacy reasons

— new datastores are being used for more specific apps

— rightscale like azure fabric controller? not really clear

— security compliance for enterprise customers? 

— clouds aren’t the best yet

— saas70?

— legacy apps?