Better Together: Scale-Out Databases on Scale-Out Storage

[clintkitson]

The introduction of scale-out persistent applications, such as databases, have changed the requirements on infrastructure. A common design pattern is to focus on local direct attached storage to satisfy storage needs. There is opportunity to transform and build a complimentary strategy for your scale-out applications with storage. Learn how to run these applications in new ways and see the possibilities that emerge.

[cduchesne]

Better Together: Scale-out databases on Scale-out storage

• Situation - Describe current best practice recommendations (local storage)
  • scale-out databases, like Cassandra, recommend dedicated local storage
  • this allows linear scale with relatively predictable performance
• Problem - Describe issues with currently recommended solution (local storage)
  • adding & removing hardware to grow/shrink a cluster, or clusters requires extra operational resources, and most importantly, time
  • creating sizeable test/dev/qa environments is costly
  • upgrades and maintenance require additional resources
  • failure of a node puts cluster at risk - time to recover this node is important - how is this recovered?
• Solution - Describe a potential solution (high performance scale-out storage)
  • containers running on "scale-out" shared storage
    • docker
    • scale-io
• Benefits - Describe benefits of scale-out storage
  • quick recovery from downed nodes due to failure or maintenance (no need to rebuild)
  • quickly deploy new clusters for dev/test/qa purposes
  • easily add additional nodes and remove unneeded nodes depending on workload demand
• Architecture - Describe different deployment architectures
  • Deployment Types
    • standard deployment
    • shared storage deployment
    • scale-io shared storage deployment
    • hyperconverged scale-io shared storage deployment
  • Scale-IO Architecture by Jeff Thomas @ Scale-IO
• Challenges - Describe some of the challenges with scale-out storage (optional)
  • how proven is solution?
  • non-standard components for cassandra world (docker, mesos, scale-io, etc)
• Demo - Ideas for possible demos below
  • Node Failure
    • show working cassandra ring w/ 5-10k tps
    • bring a node down
    • show how node comes back up on another server
    • show performance during the event
  • Grow / Shrink Cluster
    • add or remove a node
    • show ring growth/shrink
    • watch metrics while happening
  • Dev/Test/QA
    • show ease of bringing up and removing high-performance environments
• Summary - Summarize talk
  • myth about scale-out databses with shared storage
  • repeat the benefits of scale-out databases on scale-out shared storage done right

[clintkitson]

Demo idea #32