Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
The Maths of Microscaling
Search
Liz Rice
November 08, 2016
Technology
2
220
The Maths of Microscaling
Using control theory to scale containers in real time, in response to demand
Liz Rice
November 08, 2016
Tweet
Share
More Decks by Liz Rice
See All by Liz Rice
KCD Lima: eBee in Peru!
lizrice
0
140
Unleashing the kernel with eBPF
lizrice
0
250
eBPF's Abilities and Limitations: The Truth
lizrice
0
430
Simplifying multi-cloud and multi-cluster Kubernetes deployments with Cilium
lizrice
0
230
When is a Secure Connection not encrypted? And other stories
lizrice
1
91
Keeping it simple: Cilium Mesh - networking for multi-cloud Kubernetes and beyond
lizrice
1
690
How Many Proxies Do You Need
lizrice
1
160
eBPF for Security Observability
lizrice
0
1.4k
Beginner's Guide to eBPF Programming for Networking
lizrice
1
2.5k
Other Decks in Technology
See All in Technology
Large Vision Language Modelを用いた 文書画像データ化作業自動化の検証、運用 / shibuya_AI
sansan_randd
0
130
Vibe Coding Year in Review. From Karpathy to Real-World Agents by Niels Rolland, CEO Paatch
vcoisne
0
110
「使い方教えて」「事例教えて」じゃもう遅い! Microsoft 365 Copilot を触り倒そう!
taichinakamura
0
200
プロポーザルのコツ ~ Kaigi on Rails 2025 初参加で3名の登壇を実現 ~
naro143
1
190
大規模サーバーレスAPIの堅牢性・信頼性設計 〜AWSのベストプラクティスから始まる現実的制約との向き合い方〜
maimyyym
5
3.7k
Optuna DashboardにおけるPLaMo2連携機能の紹介 / PFN LLM セミナー
pfn
PRO
2
930
生成AIで「お客様の声」を ストーリーに変える 新潮流「Generative ETL」
ishikawa_satoru
1
370
AWS 잘하는 개발자 되기 - AWS 시작하기: 클라우드 개념부터 IAM까지
kimjaewook
0
120
社内報はAIにやらせよう / Let AI handle the company newsletter
saka2jp
8
1.3k
能登半島地震で見えた災害対応の課題と組織変革の重要性
ditccsugii
0
220
KMP の Swift export
kokihirokawa
0
350
OpenAI gpt-oss ファインチューニング入門
kmotohas
2
1.1k
Featured
See All Featured
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
30
2.9k
RailsConf 2023
tenderlove
30
1.2k
Making the Leap to Tech Lead
cromwellryan
135
9.6k
The Straight Up "How To Draw Better" Workshop
denniskardys
237
140k
CSS Pre-Processors: Stylus, Less & Sass
bermonpainter
358
30k
Building Adaptive Systems
keathley
43
2.8k
How STYLIGHT went responsive
nonsquared
100
5.8k
[RailsConf 2023] Rails as a piece of cake
palkan
57
5.9k
Six Lessons from altMBA
skipperchong
28
4k
Java REST API Framework Comparison - PWX 2021
mraible
33
8.9k
Build The Right Thing And Hit Your Dates
maggiecrowley
37
2.9k
The Pragmatic Product Professional
lauravandoore
36
6.9k
Transcript
The Maths of Microscaling Liz Rice @lizrice | @microscaling
What is Microscaling? Assumptions Some theory Some experiments
What is Microscaling?
Traffic spike
Too much work Spare capacity
container scaling work performance metrics
work performance metrics container scaling VM autoscaling
True for regular autoscaling too VMs take much longer to
scale
Orchestration Heterogenous services Cattle not pets
Performance targets
How many containers? Request processing time Rate of requests known?
predictable?
performance target actual performance error time t
performance target p time t actual performance x e(t) =
x(t) - p(t) e(t) → 0 error e
x(t) is proportional to n(t) n(t) = k x(t) error
e time t number of containers n
x(t) is proportional to n(t) nope! error e time t
number of containers n d(t) is proportional to e(t) d
Time delays It’s a dynamical system
Woah, the future! error e time t d(t) is proportional
to e(t + T) T d
None
Control theory!
PID controller
error e time t Proportional term d(t) = Kp e(t)
The further we are below target the more containers we need
error e time t Derivative term The faster we approach
target the fewer containers we need d(t) = Kp e(t) + Kd ė(t)
error e time t Integral term d(t) = Kp e(t)
+ Kd ė(t) + Ki e(t) Offset errors accumulated over time ∫
Which values for K? Discrete containers?
Simulator goo.gl/KAqT5y
It works! But it’s non-trivial to tune
Known behaviours Machine learning
Container parameters = metadata microbadger.com
github.com/microscaling @lizrice | @microscaling app.microscaling.com microbadger.com