Why is devops tightly coupled with the cloud

What is the cloud ?

  • what is a SaaS ?
  • what is a IaaS ?
  • what is a PaaS ?
  • We are mostly going to talk about IaaS

So how is IaaS born ?

  • Iaas was born with virtualization
  • But virtualization is not new !
  • in 1972 IBM invented hypervisors (VM/CPS 370)
    • This was the birth of virtualization
  • in 1998 VMWare was created
  • Amazon EC2 went in beta in 2006, production started late 2007

A focus on AWS

  • Virtualization API: access to computing power by an API
  • And a bit more !
  • EC2 is a server factory !
  • AWS is not only EC2
    • You probably already heard about S3
    • Other services developped over time (load balancers, databases, queues, caches, dns, mailing …)

So what’s new ?

What’s new: Technically

  • network unification (in amazon only one IP per hosts, no more juggling between IP addresses)
  • no more hardware to manage (but someone is still doing it !)
  • OS configuration streamlined, we have now identified the bare minimal
  • Easier focus on delivering applications

What’s new: Scaling

  • Can scale …
  • Scaling is also hardware (a rack is limited to 42U)
  • Scale up: unplug ip and disk, get more horse power, plug back ip and disk, running again !
  • But Scale out is better: horizontal scaling (not always easy)
  • And can also scale down (both ways we used to scale up)

What’s new: Scaling (continued)

  • Scale down, you’re thinking about long term
  • With previous concepts, but a bit further: you can scale up and down everydays
  • Jumping from capacity planning (predict money spendings) to capacity optimization (predict money savings)
  • Traffic spikes in years for your buisness (it’s christmas !)
  • This is a new way of thinking, it’s a new culture ! (“don’t repair, replace”, “don’t update, replace”)

What’s new: Availabilty

  • I can launch now, a server in asia, euope, and america, and have them all ready in 5 five minute
  • New deployement patterns with less (or no) downtime
  • Some services are already highly available (S3, ELB)
  • But don’t lie: HA is also harder with no network control

What’s new: Organization

  • This type of scaling and availabilty saves you some $ (compared to traditional methods)
  • Dev can spawn environements on demand, don’t wait for your IT service !
  • One infrastructure per project, disconnected from others infrastructures, better focus
  • Company can know how much any project costs, for integration, for prod
  • Cloud is cool for trainings !
  • A place to collaborate with customers

But cloud is not only happiness

  • No multicast
  • I/O is expensive
  • No adding of network interface
  • Beware of vendor lock-in
  • Poor customization
  • Unpredictible network performance
  • HA is also harder

Then things continued to change

  • new services (blitz.io, contiunous.io, saucelabs, cloudbees, loglly, pagerduty)
    • they were able to launch their service thanks to the cloud
    • they help you build your own service, with the cloud, for the cloud
  • more and more services
  • lot of new projects (more generic, horizontally scalable, new filesystems)
  • PaaS birth

New Challenges

  • Infrastructure aware applications
  • Roaming data stores and databases
  • Horizontal scaling needs to abstract state, do stateless where ever you can

Thanks ! Questions ?