Datacenter to AWS Cloud Migration

December 4, 2014

Theo Kim, senior director of SaaS at Jobvite describes the experience of migrating the entire company infrastructure from a dedicated data center to Amazon Web Services.

He is joined onstage by Scott Gust and Brian Morehead to describe Jobvite’s three biggest migration challenges and their solutions.


Theo Kim

  • The pain of the old data center model
  • Thinking of services rather than servers
  • Recap of gains with AWS
  • Cutting costs with reserve- and spot-instances

Scott Gust

  • Three main architectural challenges
    • uri load balancing
    • autoscaling adoption
    • network attached storage migration
  • Way to work around the elastic load balancer’s uri support
    • Used HAProxy behind ELB to do the matching
    • Important to insulate IP changes caused by ELB
  • Injecting headers with the load balancer
  • As servers become available they recreate the HAProxy config

Brian Morehead

  • The migration was fun!
  • Autoscaling made them reconsider how to automate datacenter processes
  • Moving beyond the problem of the “snowflake server”
  • Config management
    • All ec2 instances are constructed from a repo in s3
    • Chose puppet over chef or ansible
    • Standardizing hostname
    • LDAP auth
  • Puppet installs a cron job to register machines with HAProxy
  • Monitoring with Nagios and CheckMK
  • NAS migration
    • At the time of migration they had 8TB of storage and 40M+ files (resumes, attachments etc)
    • How about S3? Not really a NAS, and no metadata
    • They chose Zadara Storage to avoid stitching together EBS volumes
    • But they’re moving to an all-s3 solution
  • Copying files to s3 can be slow
    • Even multithreaded copies and bigger instances didn’t help
    • Solution: spot instances
    • Spun up 15-20 instances and it blasted the 40M files copied in three days


  • Results
  • Ended up with a 20% cost savings by moving off a managed service provider
  • Great uptime
  • Fast patches to security vulnerabilities
  • Gets to have Christmas and Thanksgiving with the family (!)
  • Ephemeral servers to do quick tests (load etc)
  • Cheap services
    • sending 275000 emails per day for $150 per month
  • Mountains of metrics from CloudWatch and CloudTrail
  • Q&A
    • Network attached storage…don’t use it!
    • Stop worrying about encryption-at-rest in the cloud
    • Data centers have a “red zone” where equipment coming out literally goes through a wood chipper
    • Intentionally undersize staging resources to force devs to make leaner code (heh heh)