This document summarizes a presentation about using Spot instances on AWS to significantly reduce costs for running mission critical applications. The key points are:
- Spotinst allows running applications on AWS Spot instances which are unused EC2 capacity available at steep discounts (up to 90% less than On-Demand prices)
- Spotinst provides solutions to overcome barriers to using Spot such as lack of capacity guarantees and need for safe application draining in case capacity is reclaimed
- The presentation includes demos of Spotinst capabilities like floating IPs, integration with AWS services like Beanstalk and CloudFormation, and a new integration with EMR autoscaling of task nodes.
- Case studies are presented from companies like Inner
2. Welcome!
To our 1st online
• Questions are always welcome
• There are no bad questions
• We’ll demo as much as we can
• Things will break, be patient
3. Amiram Shachar
Founder & CEO / Spotinst
Gal Aviv
R&D Group Manager /
Inneractive
Shannon Williams
Co-Founder & VP Marketing /
Rancher Labs
Shimon Tolts
Head of Infrastructure /
ironSource
@AmiramShachar @galaviv @shimtolts @smw355
4. Agenda
• The Basics: Spotinst on AWS
• Can I run my application on Spot?
• Demo: Spotinst floating IPs, Beanstalk, CloudFormation
• Use-case: Inneractive & Spotinst, Gal Aviv
• Use-case: ironSource & Spotinst, Shimon Tolts
• Using Rancher for Docker production workloads, Shannon
• Demo: NEW! Spotinst & EMR autoscaling
23. G a l A v i v
R & D G r o u p M a n a g e r
Spotinst Case Study
24. What we do…
• Real time trading and optimizing campaigns over ad spaces
• Real time analytics
• Machine learning and recommendations
• Trading data and insights
• Our tech stack: Java, Scala, AKKA, Spray, Kafka
Spark, Spark Streaming and Cassandra
25. Scale and numbers…
• Billions of daily incoming requests
• Tens of Billions of daily transactions
• 150ms average response time
• Hundreds of servers in dozens of elastic clusters
• Extreme scaling (up/down)
26. Scaling challenges
• Different cluster behavior (latency, CPU, load, QPS…)
• Tight back-pressure management (Scala/AKKA/Spray
stack)
• Critical serving core (99.99% uptime)
• Service cost efficiency is crucial
27. Spotinst integration
• Less than 1 day for initial setup
• Cluster by cluster partial migration
• Extreme costs reduce
• Cost vs. risk…
0%
25%
50%
75%
100%
Reserved Spotinst OnDemand
28. Nice to meet you!
Shimon Tolts
Head of infrastructure @ ironSource
30. PEOPLE REACHED EACH MONTH
APPS INSTALLED EVERY MINUTE
WITH THE IRONSOURCE PLATFORM
REGISTERED & ANALYZED DATA EVENTS
EVERY MONTH
600M 4,200 100B+
70K+
PARTNERED APPS
SEP. 2010
ESTABLISHED
3.5B+
INSTALLATIONS TO DATE
700
TOTAL EMPLOYEES
400
R&D EMPLOYEES
IRONSOURCE IN NUMBERS
TEL AVIV ISRAEL SAN FRANCISCO UNITED STATES NEW YORK UNITED STATES LONDON UNITED KINGDOM BEIJING CHINA
32. ● Our deployment - micro-services with Docker
● Implement it yourself?
● 4 Availability Zones
● 5 Auto Scaling groups? (1 on demand)
● c4.large, c3.large, r3.large, m4.large
● 4x4=16 ASG + 1 On demand
SPOT INSTANCES
33.
34. Spotinst
● One place to configure it all
● Multi instance type support
● Deep visibility into cost
● High savings
● Easy Risk/Reward mitigation (% of spot)
● Spot clean up mechanism
35. Thank you
See you in our next webinar!
@Spotinst /spotinstance/spotinst