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
Serverless Functions Azure, AWS, GCP
Search
Mikhail Shilkov
October 30, 2019
Programming
0
220
Serverless Functions Azure, AWS, GCP
Comparison of Function-as-a-Service offerings of AWS, Azure, and GCP
Mikhail Shilkov
October 30, 2019
Tweet
Share
More Decks by Mikhail Shilkov
See All by Mikhail Shilkov
From YAML to TypeScript: Developer’s View on Cloud Automation
mikhailshilkov
0
77
Cloud Management Superpowers with Pulumi
mikhailshilkov
0
480
Cloud Superpowers with Pulumi and F#
mikhailshilkov
1
470
Cloud Management Superpowers with Pulumi and .NET
mikhailshilkov
0
110
Managing Any Cloud with .NET
mikhailshilkov
0
65
Azure Infrastructure as C# and F#
mikhailshilkov
0
240
Infrastructure as Software
mikhailshilkov
0
390
Azure Infrastructure as C# and F#
mikhailshilkov
0
300
Pulumi: Cloud Infrastructure as C# and F#
mikhailshilkov
0
780
Other Decks in Programming
See All in Programming
Gemini CLIの"強み"を知る! Gemini CLIとClaude Codeを比較してみた!
kotahisafuru
3
960
React 使いじゃなくても知っておきたい教養としての React
oukayuka
18
5.5k
ワープロって実は計算機で
pepepper
2
1.2k
JetBrainsのAI機能の紹介 #jjug
yusuke
0
190
ZeroETLで始めるDynamoDBとS3の連携
afooooil
0
150
SQLアンチパターン第2版 データベースプログラミングで陥りがちな失敗とその対策 / Intro to SQL Antipatterns 2nd
twada
PRO
38
11k
可変性を制する設計: 構造と振る舞いから考える概念モデリングとその実装
a_suenami
10
1.7k
AHC051解法紹介
eijirou
0
190
Scale out your Claude Code ~自社専用Agentで10xする開発プロセス~
yukukotani
9
1.7k
PHPUnitの限界をPlaywrightで補完するテストアプローチ
yuzneri
0
390
QA x AIエコシステム段階構築作戦
osu
0
250
Strands Agents で実現する名刺解析アーキテクチャ
omiya0555
1
110
Featured
See All Featured
Designing Dashboards & Data Visualisations in Web Apps
destraynor
231
53k
Testing 201, or: Great Expectations
jmmastey
45
7.6k
Six Lessons from altMBA
skipperchong
28
3.9k
For a Future-Friendly Web
brad_frost
179
9.9k
What's in a price? How to price your products and services
michaelherold
246
12k
Why You Should Never Use an ORM
jnunemaker
PRO
58
9.5k
The Cost Of JavaScript in 2023
addyosmani
51
8.8k
Designing for Performance
lara
610
69k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
36
2.5k
GraphQLの誤解/rethinking-graphql
sonatard
71
11k
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
32
1.3k
Done Done
chrislema
185
16k
Transcript
Serverless Functions Azure, AWS, GCP Cloud Republic Event | Utrecht
| October 30, 2019
• Software developer • Cloud • Serverless • Functional programming
• Microsoft Azure MVP https://mikhail.io @MikhailShilkov Mikhail Shilkov
Function-as-a-Service Cloud Offerings
Hosting Plans Configurability Languages Programming Model Concurrency Isolation Cost Orchestrations
Cold Starts Scalability Conclusions Q&A
Hosting Plans
Azure Functions
AWS Lambda and Google Functions
Configurability
Azure Functions Consumption
AWS Lambda Memory Allocation
GCF Memory Allocation
Azure Functions Premium
Programming Languages
Language Azure Functions AWS Lambda Google Cloud Functions .NET (C#/F#)
GA GA - Node.js (JS/TS) GA GA GA JVM (Java) GA GA - Python GA (Linux only) GA GA PowerShell Preview GA - Go - GA GA Ruby - GA - Comparison Table
Programming Model
Azure Functions: Triggers and Bindings [FunctionName("MyFunc")] public static void Work(
[TimerTrigger("0 */10 * * * *")] TimerInfo timer, [Blob("filename")] Stream stream, [Queue("myqueue")] out string message) { // body }
AWS: JSON Input and Output exports.handler = async (event) =>
{ const response = { statusCode: 200, body: JSON.stringify('Hello from Lambda!'), }; return response; };
ASP.NET Core on AWS Lambda public class LambdaFunction : Amazon.Lambda.AspNetCoreServer.APIGatewayProxyFunction
{ protected override void Init(IWebHostBuilder builder) { builder .UseContentRoot(Directory.GetCurrentDirectory()) .UseStartup() .UseApiGateway(); } }
GCP: Express-like API exports.helloWorld = (req, res) => { let
message = req.query.message || req.body.message || 'Hello World!'; res.status(200).send(message); };
Deployment Artifacts
Concurrency and Isolation
Resource Pool
Resource Pool
Resource Pool
Resource Pool
Resource Pool
Resource Pool
Resource Pool
Client Value Resource Utilization
Tradeoff
Server, VM, container, process, ...
Server, VM, container, process, ... VM VM
Server, VM, container, process, ... VM VM
Server, VM, container, process, ... VM VM
Server, VM, container, process, ... VM VM
Azure Functions 12
Legacy of App Service
Legacy of App Service
App Service: scalability
App Service: scalability
Function App Consumption Plan
Isolation Layers
Simultaneous Executions
Simultaneous Executions
Simultaneous Executions
Simultaneous Executions
AWS Lambda 22
AWS Lambda
AWS Lambda
AWS Lambda
Integrating with AWS Lambda
Simultaneous Executions in AWS Lambda
Isolation Layers
Simultaneous Executions in AWS Lambda
Simultaneous Executions in AWS Lambda
Simultaneous Executions in AWS Lambda
• • • • AWS Firecracker
Isolation Layers with Firecracker
«Naive» Function Composition
Statistical Multiplexing
Google Cloud Functions 31
Google Cloud Functions
Cost
Nominal Pricing
Nuances
Azure Simultaneous Executions
AWS Lambda: HTTP Integration
AWS Lambda: HTTP Integration
Orchestration
Azure Logic Apps
AWS Step Functions
Azure Durable Functions public static async Task Sequential(DurableOrchestrationContext context) {
var conf = await context.CallActivityAsync<ConfTicket> ("BookConference", "ServerlessDays"); var flight = await context.CallActivityAsync<FlightTickets> ("BookFlight", conf.Dates); await context.CallActivityAsync("BookHotel", flight.Dates); }
Performance and Scalability
Ideal Scalability: Throughput
Ideal Scalability: Latency
Customer Value Cloud Resource Utilization
Cold Starts
Example: Loading a map
Example: Loading a map
Example: Loading a map
Cold Start
Warm Start
Period Before a Subsequent Cold Start
Cold Starts: Language Comparison
Dependencies Add to Cold Start
Fighting Cold Starts: Azure [FunctionName("Warmer")] public static void WarmUp( [TimerTrigger("0
*/10 * * * *")] TimerInfo timer) { // No need to do anything }
AWS: Single instance warming
AWS: Multiple instance warming
ETL AT SCALE: Asynchronous data processing
Async Pipelines
Experiments with Queues
AWS Lambda Processing 100k SQS Messages
GCF Processing 100k Pub/Sub Messages
Azure Functions Processing 100k Queue Messages
CPU-intensive workload on AWS Lambda
CPU-intensive workload on Google Cloud Functions
CPU-intensive workload on Azure Functions
AWS Processing Speed per Reserved RAM
HTTP AT SCALE: Serving traffic of StackOverflow
Serving StackOverflow-like traffic
Requests/sec during the load test
0 200 400 600 800 1000 1200 Azure Functions: Response
Percentiles (ms)
0 200 400 600 800 1000 1200 Azure Functions: Response
Percentiles (ms)
0 200 400 600 800 1000 1200 Azure Functions: Instances
Google Functions: Response Percentiles (ms)
Google Functions: Response Percentiles (ms)
0 200 400 600 800 1000 1200 Google Functions: Instances
0 200 400 600 800 1000 1200 AWS Lambda: Response
Percentiles (ms)
0 200 400 600 800 1000 1200 AWS Lambda: Response
Percentiles (ms)
0 200 400 600 800 1000 1200 AWS Lambda: Instances
So, What Should You Choose?
• • • • Follow-up reading (1)
• • • Follow-up reading (2)
None