Brought to you by
The Role of Machine Learning
For Cloud Native Performance
Optimization
Brian Likosar Liko
Global Director of Solutions Architecture at StormForge
Liko (lick-oh)
Global Director of Solution Architecture
■ Worked with open source for ~25 years
■ Failed statistics at University, but still appreciate P99s
■ Theme Park nerd who loves live music
■ Father, husband, friend, and mentor
K8s Resource Optimization? Sure, But How?
Some Are Already Doing It… But
Kubernetes Resource Management Complexity
PERFORMANCE
COST RELIABILITY
Container 1
CPU Memory
Requests
Limits
Requests
Limits
Replicas
Container 2
CPU Memory
Requests
Limits
Requests
Limits
Replicas
Application Settings
JVM
Heap
size
Garbage
collection
Container 3
CPU Memory
Requests
Limits
Requests
Limits
Replicas
Shuffle
file
buffer
Reducer
max
size
This Complexity Forces You to Choose…
■ Over-provisioning, driving cost
of revenues up
■ Risking application
performance & availability
issues
■ Slowing down time-to-market
so your DevOps team can
focus on achieving efficiency OVER-PROVISIONING/
COST
PERFORMANCE &
RELIABILITY ISSUES
CONFIGURATION
TIME & EFFORT
“... but this is kind of stupid, right?
Humans adapting things in YAML
files… you want to have something
automatic, AI and what not…”
(Henning Jacobs, Head of Dev Productivity, Zalando)
Key Attributes of ML-driven Optimization
In-depth app
analysis & insights
Maximize value
from existing data
Purpose-built
ML-Algorithms
Automation
(to be not stupid)
Until patterns emerge, it’s impossible to make
informed decisions
9
Until patterns emerge, it’s impossible to make
informed decisions
10
Come chat with us in person!
- KubeCon, Detroit, 24-28 Oct 2022
- AWS re:Invent, Las Vegas, 28 Nov - 2 Dec 2022
Brought to you by
Brian Likosar
liko@stormforge.io
@liko9
https://www.linkedin.com/brianlikosar

The Role of Machine Learning In Cloud Native Performance Optimization

  • 1.
    Brought to youby The Role of Machine Learning For Cloud Native Performance Optimization Brian Likosar Liko Global Director of Solutions Architecture at StormForge
  • 2.
    Liko (lick-oh) Global Directorof Solution Architecture ■ Worked with open source for ~25 years ■ Failed statistics at University, but still appreciate P99s ■ Theme Park nerd who loves live music ■ Father, husband, friend, and mentor
  • 3.
  • 4.
    Some Are AlreadyDoing It… But
  • 5.
    Kubernetes Resource ManagementComplexity PERFORMANCE COST RELIABILITY Container 1 CPU Memory Requests Limits Requests Limits Replicas Container 2 CPU Memory Requests Limits Requests Limits Replicas Application Settings JVM Heap size Garbage collection Container 3 CPU Memory Requests Limits Requests Limits Replicas Shuffle file buffer Reducer max size
  • 6.
    This Complexity ForcesYou to Choose… ■ Over-provisioning, driving cost of revenues up ■ Risking application performance & availability issues ■ Slowing down time-to-market so your DevOps team can focus on achieving efficiency OVER-PROVISIONING/ COST PERFORMANCE & RELIABILITY ISSUES CONFIGURATION TIME & EFFORT
  • 7.
    “... but thisis kind of stupid, right? Humans adapting things in YAML files… you want to have something automatic, AI and what not…” (Henning Jacobs, Head of Dev Productivity, Zalando)
  • 8.
    Key Attributes ofML-driven Optimization In-depth app analysis & insights Maximize value from existing data Purpose-built ML-Algorithms Automation (to be not stupid)
  • 9.
    Until patterns emerge,it’s impossible to make informed decisions 9
  • 10.
    Until patterns emerge,it’s impossible to make informed decisions 10
  • 11.
    Come chat withus in person! - KubeCon, Detroit, 24-28 Oct 2022 - AWS re:Invent, Las Vegas, 28 Nov - 2 Dec 2022
  • 12.
    Brought to youby Brian Likosar liko@stormforge.io @liko9 https://www.linkedin.com/brianlikosar