스타트업과 개발자를 위한 aws 클라우드 태권 세미나

Post on 06-May-2015

895 Views

Category:

Technology

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Getting to MVP (Minimum Viable Product)

Traditional World customer is known

features are known

solution is known

Traditional World

is not where we live

Most startups Know the problem, but not the solution

Many don't even know precisely what

problem they solve

Lean Startups:

LEARN & ADAPT

1. Focus on a simple implementation of your idea

1. Focus on a simple implementation of your idea

2. Start with a minimal core set of features

1. Focus on a simple implementation of your idea

2. Start with a minimal core set of features

3. Release and listen to your users

1. Focus on a simple implementation of your idea

2. Start with a minimal core set of features

3. Release and listen to your users

Minimum Viable Product

MVP Smallest thing I can do to test my idea?

a prototype shouldn't require big investments

It should be cheap and validate ideas

This Session

From 0 to MVP in 30 minutes

What matters most?

Cost of Innovation

Focus

« Want to increase innovation?

Lower the cost of failure »

Joi Ito

AWS enables you to

Fail Forward

Fail Faster

Fail Cheaper

Product Development

MVP

Time

Scale

Innovation & Iteration

MVP

Time

Scale

Innovation & Iteration

Time

Scale

Started: burbn, location-based mobile

app. Photo sharing is just one feature

Now: re-written as

photo app. Sold to FB

for 1bn

Innovation & Iteration

Time

Scale

Started: odeo, site to create & share

podcasts

Now: micro-blogging,

500M users, >10Bn

valuation

Innovation & Iteration

Time

Scale

Started: developed 51 games, none very

successful. But then game 52…

Now: raised $42M,

downloaded 1B times,

25% paid, best sold

game on AppStore

“Timing, perseverance, and ten years of trying

will eventually make you look like an overnight

success.”

Biz Stone, Twitter co-founder

AWS lowers the cost of Innovation

Time

Scale

Scenario Small team with initial idea for Mobile app

3 months to get to launch

Unknown customer/problem/solution

No cash….

Dev / Test Environment

Time

Scale

Average Spend

$0 p/m

Alpha Release

Time

Scale

Average Spend

$15 p/m

Beta Release / MVP

Time

Scale

Average Spend

$235 p/m

Getting to MVP for $250

Time

Scale

Total Spend to MVP

$250 $235 $15 $0

• 3 months dev/test/release

• Serving Beta customers

• Ready for full production

and scale

Your application

Your business & what makes you unique

Innovation, not undifferentiated heavy lifting

Spending developer time in the right place

Automate as much as you can

(Deep insight alert: Developer Time = Money)

FOCUS!

Build apps, not infrastructure

"Startups are all about focus. AWS enables focus" Ray Bradford, Kleiner Perkins, Caulfield & Byers

“Your users around the world don’t

care that you wrote your own DB”

Mike Krieger, Instagram Cofounder

AWS OpsWorks

AWS CloudFormation

AWS Elastic Beanstalk

DevOps framework

for application

lifecycle management

and automation

Templates to deploy

& manage template-

driven provisioning

Automated

resource

management – web

apps made easy

DIY / On Demand

DIY, on demand

resources: EC2, S3,

customer AMI’s, etc.

Control Convenience

Focus requires Automation

DEMO Your MVP on AWS Elastic Beanstalk

What’s AWS Elastic Beanstalk?

User Application

Application Service

HTTP Service

Language Interpreter

Operating System

Host

We Create the EC2 Instance You Focus on Developing Your App

Flexibility to Choose your Stack

We’re going to build this…

Thank You aws.amazon.com/start-ups

aws.amazon.com/ko/activate

Getting to MVP Frograms

frograms

TITLE

내 취향을 분석하는 영화추천 서비스 왓챠

어떤 영화를 볼 지 선택할 때, 취향을 분석하여 좋아할만한 영화를 추천해 주는 개인화 서비스

내 취향을 분석하는 영화추천 서비스 왓챠

베타 버전 개발 (Beta Dev Stage)

Traditional Hosting Provider

• 한대의 서버에 App서버 DB서버 함께 사용

• Loosely Coupled Architecture 필요

• Vertical Scaling (서버 스펙 업그레이드) 어려움

• Horizontal Scaling (서버 수량 증가) 어려움

• App서버 1대 DB서버 1대

• 시간 부족으로 인한 Time To Market 중요

• 사용자 유입 속도 예측 불가

• 빠르게 이용자 요구에 대처 할 수 있는 능력 필요

베타서비스 시작 (Beta Service on AWS)

App Server m1.large

DB Server m1.medium

So, We started with Simple Architecture

애플리케이션 서버 (Scale Up/Down)

AMI

App Server

m1.large m1.xlarge

• Ruby on Rails 웹 애플리케이션

• 모바일 API 개발

• 인스턴스 타입 변경이 수분안에

가능

• AMI (Amazon Machine Image)를

통한 인스턴스 타입 업그레이드

Right Instance Type

애플리케이션 서버 (Customized AMI)

애플리케이션 서버 (Rapidly Scale Out/In)

Elastic Load Balancer

App Servers

With • 순간적으로 폭증하는 트래픽

• SES를 통한 대량 Email 발송처리

• 수분만에 Spot Instance 추가

• 트래픽 처리 끝난 후 모두 제거

Marketing Promotion

m3.xlarge

데이터 베이스 (Test & Apply)

Extra Large Large Medium

• 실시간 Modify 를 통한 무중단 업그레이드

App Servers

• 초창기 단순 구조, Read Replica 1-click 추가

또한, 그 외에도…

Application Server

Database Server

추천서버 Cache 서버 Search 서버

m3.xlarge On-demand

Spot Instance

Extra Large Read Replica

M1.xlarge

추천 관련된 모든 역할을 담당하는 서버 DB, 캐시서버, 어플리케이션 서버와 통신

m1.xlarge

Redis, Memcached

m1.xlarge

Sphinx

Next Step with AWS

“앞으로 개개인에게 특화된 '개인화' 서비스에 미래가 있다고 생각해요. 영화 뿐만 아니라 개인화 할 수 있는 것들은 무궁무진합니다.

이러한 분야에서 빠른 시장진입을 위해서 AWS가 많은 도움을 줄것입니다.”

드라마

영화

게임 뉴스

음악

문화컨텐츠

동영상

쇼핑

공연

맛집

Innovators Needed !!

경험 많은 서버/백엔드 개발자

Machine Learning 전공자

contact@frograms.com

Getting to Scale

503 Service Temporarily Unavailable

The server is temporarily unable

to service your request due to

maintenance downtime or capacity

problems. Please try again later.

With AWS, scale from one instance…

…to thousands

Fully automated!

BUT…

How do I scale my architecture to

support my first 10M users?

“Think Big, Start Small, Scale Fast”

Eric Ries, author of NY Times

bestseller “The Lean Startup”

01 02 03 04

Idea MVP Profitability Scale

Getting to Scale

By building a scalable Architecture to

support your first 10M users

1. Dev & Test

2. Alpha Release

3. Beta Release

Production 1.0

Architecture

Database Options

Self-Managed Fully-Managed

Database Server

on Amazon EC2

Your choice of

database running on

Amazon EC2

Bring Your Own

License (BYOL)

Amazon

DynamoDB

Managed NoSQL

database service

using SSD storage

Seamless scalability

Zero administration

Amazon RDS

Relational Database

as a managed

service

Flexible licensing:

BYOL or License

Included

But how do I choose what

DB technology I need?

SQL? NoSQL?

Some folks won’t like this.

But…

Start with SQL databases

But, but, but, but…

No. You don’t.

Start with SQL databases

Established and well worn technology

Lots of existing code, communities, books, tools, etc

Clear patterns to scalability

You aren’t going to break SQL DBs in your first 10 million users. No really, you won’t

Why SQL?

• Database-as-a-Service

• No need to install or manage database

instances

• Scalable and fault tolerant configurations

Feature Details

Platform support Create MySQL, SQL Server and Oracle

Preconfigured Get started instantly with sensible default settings

Automated patching Keep your database platform up to date automatically

Backups Automatic backups and point in time recovery using snapshots Manual DB snapshots

Failover Automated failover to slave hosts in event of a failure

Replication Easily create read-replicas of your data and seamlessly replicate data across availability zones

Amazon Relational Database Service (RDS)

Automatic resizing of

compute clusters based on

demand Trigger auto-scaling policy

Feature Details

Control Define minimum and maximum instance pool sizes and when scaling and cool down occurs.

Integrated to Amazon CloudWatch

Use metrics gathered by CloudWatch to drive scaling.

Instance types Run Auto Scaling for On-Demand and Spot Instances. Compatible with VPC.

as-create-auto-scaling-group MyGroup

--launch-configuration MyConfig

--availability-zones us-east-1a

--min-size 4

--max-size 200

Auto-Scaling Amazon CloudWatch

Production 1.0 Architecture

Production 1.0 Architecture

Well-designed, 2 Tier architecture

Highly Available due to Multiple Availability Zone

Load Balancing & Auto-Scaling for full scalability

Fully managed Database included

Capable of serving >10K-100Ks users

BUT…

Production 1.0 Architecture

Wasted server capacity for static content

Reliability and durability are not yet optimal

End-user experience could be improved thru offloading & caching

SO…

Let’s add

Simple Storage Service (S3)

CloudFront to optimize the end-user experience

Durable storage, any object

99.999999999% durability of objects

Unlimited storage of objects of any type

Up to 5TB size per object

Feature Details

Flexible object store Buckets act like drives, folder structures within

Access control Granular control over object permissions

Server-side encryption 256bit AES encryption of objects

Multi-part uploads Improved throughput & control

Object versioning Archive old objects and version new ones

Object expiry Automatically remove old objects

Access logging Full audit log of bucket/object actions

Web content hosting Serve content as web site with built in page handling

Notifications Receive notifications on key events

Import/Export Physical device import/export service

Simple Storage Service (S3)

• World-wide content distribution

network

• Easily distribute content to end

users with low latency, high data

transfer speeds, and no

commitments Feature Details

Fast Multiple world-wide edge locations to serve content as close to your users as possible

Integrated with other services Works seamlessly with S3 and EC2 origin servers

Dynamic content Supports static and dynamic content from origin servers

Streaming Supports rtmp from S3 and includes support for live streaming from Adobe FMS and Microsoft Media Server

CloudFront

Production 1.2

Architecture

Production 1.2 Architecture

Well-designed, 2 Tier architecture

Highly Available due to Multiple Availability Zone

Load Balancing & Auto-Scaling for full scalability

Fully managed Database included

Static content stored in durable, consistent way

Improved end-user experience through CDN

Capable of serving >100K-1M+ users

BUT…

Production 1.2 Architecture

You are now at Scale…

…with lots of data…

…and need to optimize continuously.

But how and where?

SO…

Let’s add

Big Data for analytics of web, mobile, gaming,

and log data

Multiple managed AWS services for Big Data

• Managed, elastic Hadoop cluster

• Integrates with S3 & DynamoDB

• Leverage Hive & Pig analytics scripts

Feature Details

Scalable Use as many or as few compute instances running Hadoop as you want. Modify the number of instances while your job flow is running

Integrated with other services

Works seamlessly with S3 as origin and output. Integrates with DynamoDB

Comprehensive Supports languages such as Hive and Pig for defining analytics, and allows complex definitions in Cascading, Java, Ruby, Perl, Python, PHP, R, or C++

Cost effective Works with Spot instance types

Monitoring Monitor job flows from with the management console

Elastic MapReduce (EMR)

Foursquare… Founded in 2009 112M in Venture Capital 33 million users 1.3 million businesses using the service

…generates a lot of Data 3.5 billion check-ins 15M+ venues, Terabytes of log data

Uses EMR for Evaluation of new features

Machine learning

Exploratory analysis

Daily customer usage reporting

Long-term trend analysis

Benefits of EMR

Ease-of-Use “We have decreased the processing time for urgent data-analysis”

Flexibility To deal with changing requirements & dynamically expand reporting clusters

Costs “We have reduced our analytics costs by over 50%”

Production 1.3

Architecture

Production 1.3 Architecture

Well-designed, 2 Tier architecture

Highly Available due to Multiple Availability Zone

Load Balancing & Auto-Scaling for full scalability

Static content stored in durable, consistent way

Improved end-user experience through CDN

Big Data analytics built in for continuous optimization

Capable of serving >1m-10M+ users

DEMO Getting to Scale

Thank You aws.amazon.com/start-ups

aws.amazon.com/ko/activate

Getting to Scale

2011. 9

2013. 9

Architecture on AWS

S3 Bucket

ELB Amazon SES Worker

Database

Client

CloudFront

Search Engine AMIs Route53 Monitoring ElastiCache

Web

MQ

Elastic Compute Cloud

Elastic Compute Cloud

~5 Instances

~32 Instances

Elastic Load Balancer

HTTPS Request HTTP

CloudFront

CloudFront

ElastiCache

ElastiCache

Getting to Profitability

Time

Usage

Page Views

Revenue

Etc.

The Infamous Hockey Stick

Time

Usage

Page Views

Revenue

Etc.

The Infamous Hockey Stick

Costs

You want only 3 things

Revenue to go Up

Unit Costs to go Down

Margin to go Up

Time

Usage

Page Views

Revenue

Etc.

The Infamous Hockey Stick

Costs

How does AWS help here?

Economies of Scale

Pricing Models

Cost Aware Architecting

What does this look like in

the real world?

An example

Enterprise software provider in APAC

Focused on SaaS for storage, security, collaboration, etc.

Backed by leading VC’s in the region

Strong growth – winning customers globally

Focused on profitability & reducing unit costs

Worked closely with the AWS team to optimize its architecture

Margin

Growth

-10%

price drop

in S3

-20%

RI purchase

-22%

Migration

Cassandra

to Dynamo

-18%

Price drop in

S3 of 25%

54%

reduction in

unit costs

“Based on a True Story”

01 02 03 04

Idea MVP Profitability Scale

Getting to Profitability 03 04

Profitability Scale

Pricing Models

Cost Aware Architecting

Total Cost of Ownership

On-Demand

Pay for compute

capacity by the hour

with no long-term

commitments

For spiky workloads,

or to define needs

Cost Optimization using different purchase models

Reserved

Make a low, one-time

payment and receive a

significant discount on

the hourly charge

For committed

utilization

Spot

Bid for unused capacity,

charged at a Spot Price

which fluctuates based

on supply and demand

For time-insensitive or

transient workloads

Free Tier

Get Started on AWS

with free usage & no

commitment

For POCs and

getting started

aws.amazon.com/ko/activate

Reserved Instance Pricing

Make a low, one-time payment and receive a

significant discount on the hourly charge

For committed utilization

•Light Utilization RI

•Medium Utilization RI

•High Utilization RI

•1-year

•3-year

2 Terms 3 Versions

Reserved Instance Pricing

Utilization RI option Savings over On-Demand

<10% On-Demand

10% - 40% Light Utilization RI Up to 56%

40% - 75% Medium Utilization RI Up to 66%

>75% Heavy Utilization RI Up to 71%

• Most traffic happens in the afternoons and evenings, so they reduce the number of

instances at night by 40%.

• At peak traffic $52 an hour is spent on EC2 and at night, during off peak, the spend is as

little as $15 an hour. Saving per hour = 71%

Save more money by using Spot Instances

Up to 85% savings over On Demand pricing

Spot market for under-utilized capacity Requested Bid Price and Pay as you go

Spot Price < On-Demand Price

Use Case Types of Applications

Batch Processing Generic background processing (scale out computing)

Hadoop Hadoop/MapReduce processing type jobs (e.g. Search, Big Data, etc.)

Scientific Computing Scientific trials/simulations/analysis in chemistry, physics, and biology

Video and Image Processing/Rendering

Transform videos into specific formats

Testing Provide testing of software, web sites, etc

Web/Data Crawling Analyzing data and processing it

Financial Hedgefund analytics, energy trading, etc

HPC Utilize HPC servers to do embarrassingly parallel jobs

Cheap Compute Backend servers for Facebook games

Use Cases for Spot Pricing

Optimizing Video Transcoding Workloads

for a FREEMIUM model

Free Offering Optimize for reducing cost

Acceptable Delay Limits

Implementation – Leverage spot pricing

– Maximum Bid Price

– < On-demand Rate

– Use on-demand Instances, if delay

Get strongly reduced price for your workload

Premium Offering Optimized for Faster response

No Delays

Implementation

– Invest in Reserved Instances

– Use on-demand for Elasticity

Get Instant Capacity for higher price

Getting to Profitability 03 04

Profitability Scale

Pricing Models

Cost Aware Architecting

Total Cost of Ownership

“Give me 4 fault tolerant algorithms and I can pick

the best one almost with my eyes closed.

If you then ask me which one is best for the

business, in terms of dollar costs, I would be

clueless...”

Werner Vogels, CTO, Amazon

Cost optimization through ‘Cost Aware Architecting’

…by leveraging: Reduce Cost of…

Compute 1. S3 & CloudFront for Caching & Offloading

Storage 3. Storing derivative objects in S3 ‘Reduced Redundancy’

Database 4. Read Replicas and/or ElastiCache

Test & Dev 5. Rapid proto-typing & Lean Dev/Test

2. Auto-Scaling done Right

1. S3 & CloudFront for Caching & Offloading

• Reduce your compute demand and costs

• Improve end-user experience

• Increase reliability and durability

Cost Aware Architecting to Reduce costs of EC2

1. S3 & CloudFront for Caching & Offloading

Cost Aware Architecting to Reduce costs of EC2

1. S3 & CloudFront for Caching & Offloading

Cost Aware Architecting to Reduce costs of EC2

1. S3 & CloudFront for Caching & Offloading

Cost Aware Architecting to Reduce costs of EC2

1. S3 & CloudFront for Caching & Offloading

Cost Aware Architecting to Reduce costs of EC2

2. Auto-Scaling done Right with Real Time reaction response

• Elastic Load Balancing and (event-driven) Auto Scaling

• Notification of pending news flash (with audible alarm)

• On-demand ramp up of capacity (6 mins.)

• Subscriber alert push delivered

• Mass response traffic handled (followed by ramp down)

Cost Aware Architecting to Reduce costs of EC2

Buuuk for Singapore Press Holding (SPH)

2. Auto-Scaling done Right with Real Time reaction response

Cost Aware Architecting to Reduce costs of EC2

Straits Times Buuuk

2. Auto-Scaling done Right with Real Time reaction response

Cost Aware Architecting to Reduce costs of EC2

2. Auto-Scaling done Right with Real Time reaction response

Cost Aware Architecting to Reduce costs of EC2

2. Auto-Scaling done Right with Real Time reaction response

Cost Aware Architecting to Reduce costs of EC2

2. Auto-Scaling done Right with Real Time reaction response

Cost Aware Architecting to Reduce costs of EC2

3. Storing derivative objects in S3 ‘Reduced Redundancy’

• Original vs. derived assets : 33% savings

• Single reference and consistency

• Control, accurate logs and tracking

Cost Aware Architecting to Reduce costs of S3

Reduced Redundancy Storage

‘RRS’

4. Read Replicas and/or ElastiCache (‘Database Smarts’)

• Scale out and share work

• Optimal performance, minimize load

• Enhance reliability, ensure data safety

• Cost reduction

Cost Aware Architecting to Reduce costs of DB

5. Rapid proto-typing & Lean Dev/Test

• Inexpensive idea validation

• Seamless switch over and versioning

• Rapid dev / test agility

Cost Aware Architecting to Reduce costs of Test/Dev

Getting to Profitability 03 04

Profitability Scale

Pricing Models

Cost Aware Architecting

Total Cost of Ownership

When calculating TCO…

#1 Start by understanding your use cases & usage patterns

Traditional HW / Hosting

On and Off Fast Growth

Predictable peaks Variable peaks

WASTE

CUSTOMER DISSATISFACTION

AWS = Elastic Capacity

Fast Growth On and Off

Predictable peaks Variable peaks

When calculating TCO…

#1 Start by understanding your use cases & usage patterns

#2 Apples to Apples – Take all the fixed costs into consideration

When calculating TCO…

#1 Start by understanding your use cases & usage patterns

#2 Apples to Apples – Take all the fixed costs into consideration

#3 Leverage ‘Cost Aware Architecting’ to reduce resources

0

10

20

30

40

50

60

Hosting

Traditional Hosting vs AWS

# of

(virtual)

servers

Offload

to S3

Caching

with CF

Auto-

Scaling Etc. Hosting

When calculating TCO…

#1 Start by understanding your use cases & usage patterns

#2 Apples to Apples – Take all the fixed costs into consideration

#3 Leverage ‘Cost Aware Architecting’ to reduce resources

#4 Include pricing models (RI, Spot) and economies of scale

Margin

Growth

-10%

price drop

in S3

-20%

RI purchase

-22%

Migration

Cassandra

to Dynamo

-18%

Price drop in

S3 of 25%

54%

reduction in

unit costs

“Based on a True Story”

When calculating TCO…

#1 Start by understanding your use cases & usage patterns

#2 Apples to Apples – Take all the fixed costs into consideration

#3 Leverage ‘Cost Aware Architecting’ to reduce resources

#4 Include pricing models (RI, Spot) and economies of scale

#5 Take a look at what’s included: Intangible Cost Savings !

New Customers Amazon EC2

Amazon RDS

Amazon ELB

Amazon S3

Amazon EBS

For All Customers Amazon SQS/SNS

Amazon DynamoDB

Amazon SES

Amazon SWF

And more…

AWS Elastic Beanstalk

AWS CloudFormation

AWS IAM

Auto Scaling

Consolidated Billing

No Charge for

Inbound Data Transfer

Data Transfer between

Instances within an

Availability Zone

Free Usage Tier

Did you know?

Free Services Data Transfer

Trusted Advisor

A premium security spec at non-premium

prices

• Security groups for EC2

and VPC

Network ACL

• Multi-Factor Authentication

• CloudHSM

• RDS Oracle transparent

encryption

• VPC

• Direct connect

• Dedicated instances

• Identity & Access

Management

• S3 Encryption

DEMO Getting to Scale

Off-loading of static content to CloudFront to

reduce required server capacity

So what does this mean in terms of costs?

Month

Medium EC2 instances 1 $ 121

CloudFront Data Transfer Out 1Tb $ 168

CloudFront Requests $1.89

TOTAL $ 291

Month

Medium EC2 instances 4 $ 485

AWS Data Transfer Out 1Tb $ 194

TOTAL $ 679

Standard Architecture Optimized Architecture

57% lower cost – 6 x faster

Thank You aws.amazon.com/start-ups

aws.amazon.com/ko/activate

Getting to Profitability on AWS

the beatpacking company

초기

투자비용

인력

안정성

스타트업 인프라 선택의 고려요소

MONEY

AWS 비용 구조

대부분의 경우 가장 많은 부분을 차지

Amazon

EC2

EC2 절약의 2가지 키워드

24 시간 내내 같은 양의 인스턴스를 사용할 필요가 없다.

EC2 절약의 2가지 키워드

On-Demand와 멀어질 수록 비용은 최적화 된다.

AWS EC2 Instance Types

On-Demand

• Pay As You Go • 원하는 언제든 생성/삭제 가능 • 시간당 가장 비싸게 이용

AWS EC2 Instance Types

Reserved

• 1년 혹은 3년의 기간 + 사용 정도로 미리 약정 • On-Demand 대비 저렴한 시간당 요금 • 계약하는 만큼 언제든 이용 가능 • 초기 선결제 금액

AWS 가용영역에 존재하는 여유 자원을 경매를 통해 이용

인스턴스 생명 주기와 수량을 통제할 수 없음

EC2 Spot Instance

EC2 Spot Instance

하지만, 잘만 활용하면 최대 80% 정도 비용 감축 가능

요금 비교

시간당(달러)

시간당(원) 할인율

On-Demand $0.740 \792.48

Reserved/1yr $0.511 \547.49 -31%

Reserved/3yr $0.358 \383.56 -51%

Spot $0.192 \205.71 -74%

AWS Tokyo Region / 1a / c1.xlarge / 2013-10-12 환율 기준 Reserved Heavy Utilization, 선결제 금액을 시간당 환산 반영

EC2 Spot Instance

c1.xlarge / us-east

EC2 Spot Instance

c1.xlarge / us-east

EC2 Spot Instance

• 가용영역에 따라 서로 다른 가격 • 가용영역에 따른 서로 다른 패턴

• 반면 어떤 가용영역은 계속 안정된 가격 흐름을 보인다

• 주된 가용영역 선택에 있어 고려 요소

EC2 Spot Instance – 입찰 전략

• On-Demand 가격에 입찰하여 가격 통제

• 그 시각 최저가로 낙찰되므로 무조건 저액을 써낼 필요는 없음

EC2 Spot Instance

c1.xlarge / ap-northeast-1

EC2 Spot Instance

Spot의 단점인 인스턴스 생명주기를 관리하지 못하는 점 AutoScaling 전략을 통해 극복

Auto Scaling

EC2 Spot Instance + AutoScaling

On-Demand Group Spot Group

• 부하 증가에 천천히 반응하여 Scale-Up

• 부하 감소에 빠르게 반응하여 Scale-Down

• 부하 증가에 빠르게 반응하여 Scale-Up

• 부하 감소에 천천히 반응하여 Scale-Down

Spot 인스턴스 위주로 부하를 수용하고, Spot 인스턴스 부족분을 On-Demand 로 수용

EC2 Spot Instance + AutoScaling

6:00 9:00 12:00 15:00 18:00 21:00 0:00

On-Demand Spot Load

Spot : 5분간 평균 CPU이용률 40% OD : 5분간 평균 CPU이용률 60%

EC2 Spot Instance + AutoScaling

6:00 9:00 12:00 15:00 18:00 21:00 0:00

On-Demand Spot Load

Spot 확보에 실패한 경우 On-Demand가 Scale-Up

Spot 확보에 따른 On-Demand Scale-Down

EC2 Spot Instance + AutoScaling

301.92

68.448 59.144

0

50

100

150

200

250

300

350

On-Demand On-Demand + Spot RI(1yr) + Spot

BEAT API 서버 비용 비교

일 비용($)

77%

( 세줄)정리

• Spot 인스턴스를 최대한 활용하여 비용 절감 • Spot 인스턴스의 단점을 AutoScaling 전략으로 커버 • 어느 정도 부하 패턴이 일정해 지면

On-Demand 인스턴스를 RI 인스턴스로 변경

감사합니다!

kkung@beatpacking.com

Q&A

top related