Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Can AI Outperform Engineers in Optimizing Syste...

Stefano Doni
April 29, 2024
18

Can AI Outperform Engineers in Optimizing System Performance and Efficiency?

Can AI outperform engineers in optimizing system performance and efficiency?

Those is one of my favourite questions I got asked during the recent webinar with Sabre.

I prepared a quick deck showing 3 super interesting and counter-intuitive examples. We used Akamas to optimzie different systems, and we were suprised looking at the optimal configuration AI identified. As performance engineers, we would have never tried such configurations... but it turned out, they provided double digit cost or performance improvements!

#1: Improving Java application performance via JVM tuning
#2: Reducing costs of an app running on EC2 by picking optimal instance types
#3: Optimizing throughput of a MongoDB database via MongoDB+OS tuning

Stefano Doni

April 29, 2024
Tweet

More Decks by Stefano Doni

Transcript

  1. © 2024 Akamas • All Rights Reserved • Confidential Can

    AI Outperform Engineers in Optimizing System Performance and Efficiency? Stefano Doni, Akamas CTO 1
  2. © 2024 Akamas • All Rights Reserved • Confidential Example

    1# Optimizing Java application performance 2
  3. © 2024 Akamas • All Rights Reserved • Confidential 3

    Want faster Java apps? Reduce garbage collection pause is the mantra everyone follows https://www.akamas.io/resources/beware-improving-gc-performance-may-lead-to-slower-applications
  4. © 2024 Akamas • All Rights Reserved • Confidential 4

    We asked AI to make a Java app faster via JVM tuning. AI reduced latency by 29%... while INCREASING GC time https://www.akamas.io/resources/beware-improving-gc-performance-may-lead-to-slower-applications
  5. © 2024 Akamas • All Rights Reserved • Confidential Example

    2# Optimizing cost efficiency of an application on EC2 5
  6. © 2024 Akamas • All Rights Reserved • Confidential 6

    Want cheaper EC2 apps? Leverage 100s of cloud infrastructure options https://speakerdeck.com/stefanodoni/a-needle-in-the-haystack-optimizing-cloud-configurations-for-price-performance
  7. © 2024 Akamas • All Rights Reserved • Confidential AI

    found a counter-intuitive instance type that provided -24% costs and+12% throughput 7 https://speakerdeck.com/stefanodoni/a-needle-in-the-haystack-optimizing-cloud-configurations-for-price-performance
  8. © 2024 Akamas • All Rights Reserved • Confidential Example

    3# Optimizing latency of a MongoDB database 8
  9. © 2024 Akamas • All Rights Reserved • Confidential 9

    AI identified optimal full-stack configurations (MongoDB+OS) leading to 2x query/sec https://speakerdeck.com/stefanodoni/achieve-superhuman-performance-with-machine-learning
  10. © 2024 Akamas • All Rights Reserved • Confidential 10

    Optimal OS settings found by AI apparently made disk IO slower… but that made MongoDB faster https://speakerdeck.com/stefanodoni/achieve-superhuman-performance-with-machine-learning
  11. © 2024 Akamas • All Rights Reserved • Confidential 11

    akamas.io/free-trial See for yourself, try Akamas free trial!
  12. Contacts [email protected] @AkamasLabs @akamaslabs Italy HQ Via Schiaffino 11 Milan,

    20158 +39-02-4951-7001 USA East 211 Congress Street Boston, MA 02110 +1-617-936-0212 Singapore 5 Temasek Blvd Singapore 038985 USA West 12130 Millennium Drive Los Angeles, CA 90094 +1-323-524-0524 LinkedIn Twitter Email © 2024 Akamas • All Rights Reserved • Confidential 12