Previous proposalUsing bcache to get a performing persistent storage in AWS
Next proposalBlueprinting your Infrastructure with Ansible
Getting bare metal performance with AWS
Submitted by Anshu Prateek (@anshprat) on Wednesday, 25 March 2015
Section: Full talk Technical level: Intermediate Status: Rejected
To discuss best performance optimization techniques with AWS EC2
Virtualization has long been considerd as the way to go if you need a quick, cheap and easily available resource. But the performance in virtualization world, at bests, is a best effort service. Amazon has, in the last one year, changed the ball game to the extent where the performance on the EC2 instances have become at par with their bare metal counterparts. Advent of technologies like HVM, VPC, Enhanced Networking, Placement group, ENI have helped Aerospike to attain a Million TPS on a single node. So far this distinction was feasible only on bare metals.
These performance enrichments and tunings are application agnostic and can be used by any tenant on the AWS stack.
Basic knowledge of AWS terms and techniques.
I have been working in the domain of Large Scale Production Engineering for the last six years. Starting with Yahoo! Search, and having touched all the 1000+ machines in the video search world, I moved on to the niche world of NoSQL Databases with the leader of performance pack, Aerospike. Coming from a world of 700ms SLA, the world of a sub-millisecond latency SLA was like jumping into an F1 race after driving monster trucks. At Aerospike, other than tooling and product, I worked on the performance benchmarking and enhancement efforts for Aerospike on AWS and other cloud providers.