Running in the Cloud: A Story Justin Baugh Audeo, Inc. firstname.lastname@example.org www.audeocloud.com
The Client: PUMA• Global sportswear and apparel company – $3bn a year in sales• PUMA digital marketing strategy executed by PUMA Interactive – Huge leeway in decision making – Regional team execution of global strategy• Global IT is “hands off” – Websites? That’s marketing!
The Old Way• Years of organic growth – No strategy or central authority – Agencies come and go - what they deliver is forever• Lack of IT operations experience – Entirely reactive• Fragmented relationships – Dozens of microsites – Confusing user experience – Cornucopia of acronyms – FIVE different vendor platforms
Pain Points• Lack of metrics – No clear goals• Cost effectiveness – Multiple complex relationships• Finger pointing – Who, ultimately, fixes things?• Lack of cohesiveness & unified branding – Subdomains, separate domains, etc – Is this a PUMA site?
The Solution• Reimplementation of puma.com – Ruby on Rails CMS – Leverage enterprise systems via custom middleware – Fully internationalized – Everything is a subsite – Unified branding, navigation, search• Phase out all other vendors – Redeploy on a private cloud – One vendor, one contact point!
Why a Cloud?• Abstraction – Computing as definable utility – Deploy resources when needed – Scale with ease• Cost Efficiencies (CapEx / OpEx) – Pay for what you use – Scale when necessary! – Commodity hardware – Uniform platforms & solutions
Why Eucalyptus?• Control your own destiny – Public clouds • performance variability • black box • unknown SPOFs – When it’s down: “who you gonna call?” – How much downtime is acceptable? – Leverage existing resources• Full control over your “building blocks” – You determine RAM, CPU, SAN, etc – Custom performance envelope – Load balancers, failover solutions, multisite, etc
…Sounds expensive!• It Depends! – How are you implementing it? – What is your current environment? • Managed hosting? • Reusing existing hardware? • Buying new hardware?• Illustrating cost savings / ROI: – Utilities have a known per hour cost – Demonstrate this cost – and beat it! – Don’t cheat: OpEx / CapEx over time – Other demonstrable benefits: • scalability, reusability, reliability
Case Study: PUMA• Two availability zones – 5 NCs in each zone (40 cores / 160GB RAM) – Upgraded to 480GB in 2012 – NetApp dualhead SAN (EBS/NFS) – F5 load balancer pair – Failover machines in case of cloud / hardware failure• Consolidation / efficiency benefits – Entire setup costs about the same as four machines with Vendor X• Cost benefit vs. public clouds – 30-50% savings over EC2 / Rackspace• Technology benefits – Support a variety of technologies and platforms in a cohesive manner – Redundancy built-in
Okay, I’m in! Now what?• This stuff is magic, right? It Just Works™! – Old tricks aren’t necessarily valid in the cloud • Ephemeral resources • Availability (and safety) not guaranteed • Machines are not monoliths – New processes are needed • Flexibility, scalability enabled by reproducibility • Your technology does not magically scale – Puppet, Chef, etc are your friends – Documented (shriek!) and expressible processes – Your application should ideally be distributable AND fault tolerant!
Conclusion• The cloud is not going away – If you’re not using it now, your competitors are – Enables global reach, negligible startup costs• The benefits are demonstrable, however: – Don’t assume savings: prove it! – Development & operations must leverage benefits – Don’t oversell: this is not magic