Amazon EC2 in der Praxis

Amazon EC2 in der Praxis

Jonathan Weiss

Scalarium

Amazon EC2 Cluster Management

  Auto-Config

  Self-Healing

  Auto-Scaling

  One-click-deployment

www.scalarium.com

2

3

Amazon Elastic Compute Cloud

Virtual Server via API call

Pay per hour

Different instance types in different regions

4

EC2 Features

Spot instances

Elastic IP

Elastic Load Balancing

CloudWatch

Auto Scaling

Virtual Private Cloud 

Elastic Block Storage

5

6

Running on EC2 Setting up an app

Running on EC2

Steps to complete

  Choose region and availability zone

  Define architecture

  Choose instance types and AMI

  Boot & configure instances

  Deploy application

  …

  Profit

7

Region and Availability Zone

Regions   Individual EC2 installations

  Multiple datacenters a.k.a availability zones

  Small differences in pricing

  US-east usually gets features first

Availability Zone

  Datacenter of instances & volumes

  Differently named for different accounts

  Traffic within the same zone is free

  Different Instance type availability

8

Existing Availability Zones

US-EAST-1

  us-east-1a

  us-east-1b

  us-east-1c

  us-east-1d

9

US-WEST-1

  us-west-1a

  us-west-1b

EU-WEST-1

  eu-west-1a

  eu-west-1b

AP-SOUTHEAST-1

  ap-southeast-1a

  ap-southeast-1b

Single AZ Deployment

Most common and simple setup

Not very different from typical setup

10

Multi AZ Deployment

11

Architecture

12

Architecture

13

Architecture

14

Elastic IP

  Static, public IP in contrast to instance IP

 Associate to any instance (~15s)

  Paying when not in use

 Use for DNS of load balancer

15

Elastic Block Store EBS

 Network based block device: 1GB – 1TB

  Persistent & redundant

  Can be used by one instance at a time

  Incremental Snapshots

  RAID for better performance

16

Elastic Load Balancer ELB

 Managed Load Balancer

 Auto-Scales via Round Robin DNS

  Can start/stop instances

  Slower and not as flexible as instance + HAProxy

The only way to scale past 1 Gbit/s!

17

Relational Data Store RDS

 Managed MySQL instance

  Choose instance size,

pay ~ 30% premium

 Hot-Failover and read replication

available

 Define a 4h maintenance window

  Better performance than single EBS-backed instance

18

Instance Types

19

RAM Cores x ECU Arch Hourly Cost (US-East)

Monthly Cost (US-East)

Micro

Tiny 613 MB Burst 1 x 2 32bit/64bit $0.02 $15

Standard

Small 1.7 GB 1 x 1 32bit $0.085 $62

Large 7.5 GB 2 x 2 64bit $0.34 $248

Extra Large 15 GB 4 x 2 64bit $0.68 $496

High-Memory

Extra Large 17.1 GB 2 x 3.25 64bit $0.50 $366

Double Extra Large 34.2 GB 4 x 3.25 64bit $1.20 $878

Quadruple Extra Large 68.4 GB 8 x 3.25 64bit $2.40 $1756

High-CPU

Medium 1.7 GB 2 x 2.5 32bit $0.17 $124

Extra Large 7 GB 8 x 2.5 64bit $0.68 $496

1 ECU = 1.0 – 1.2 GHz Opteron/Xeon from 2007

Recommendation

Start with   Small for load balancer

  Medium for App server

  Large for DB server

  Tiny for monitoring and utility

Grow to

  XL-HighCPU for load balancer

  XL-HighCPU for App server

  XL-HighMemory for DB server

20

Reserved Instances

Pay amount upfront and you get:

Hourly price ~ 33%

Guaranteed availability

24/7 instance cost: ~ 55%

Recommended!

21

Running on EC2

Steps to complete

 Choose region and availability zone

 Define architecture

 Choose instance types

  Boot & configure instances

  Deploy application

  …

  Profit

22

23

Instance Configuration From blank Linux to your Java/Rails/PHP app server

Instance Bootstrap

24

2 Approaches

Instance Bootstrap

Images

  One image for every server type

  All software pre-installed

  Few moving parts

  Fast boot

25

2 Approaches

Instance Bootstrap

Images

  One image for every server type

  All software pre-installed

  Few moving parts

  Fast boot

26

Scripted

  One image to rule them all

  Smallest common denominator

  Acquires configuration on start

  Installs & configures during boot process to fit role

  All parts moving

  Slower boot

2 Approaches

Instance Bootstrap

Images

  One image for every server type

  All software pre-installed

  Few moving parts

  Fast boot

27

Scripted

  One image to rule them all

  Smallest common denominator

  Acquires configuration on start

  Installs & configures during boot process to fit role

  All parts moving

  Slow boot

2 Approaches

Chef

Open Source Provisioning Tool

  Configures your servers

  Cookbooks & recipes

  Ruby DSL

28

Chef

29

Chef Solo

30

MemcachedSolo

31

Chef Solo Run

32

Configuration JSON

Run Chef Solo

Deploying

Deploy resource   Compatible with Capistrano layout

  Supports Capistrano hooks via deploy/hook_name.rb

  Async!

33

Chef Experiences

Great so far, but

  Not idempotent

  Unintuitive 2 phases

  Documentation could be better

  Example Cookbooks often so so

34

Running on EC2

Steps to complete

 Choose region and availability zone

 Define architecture

 Choose instance types

 Boot & configure instances

 Deploy application

  …

  Profit

35

The Project

Browser game hosted on EC2

1 million DailyActive Users (DAU)

~ 130 servers

~ 100.000 requests per minute on average

Flexibility and Scalability

Flexibility and Scalability

© Peritor GmbH - Alle Rechte vorbehalten

Peritor GmbH Blücherstr. 22, Hof III Aufgang 6 10961 Berlin

Tel.: +49 (0)30 69 20 09 84 0 Fax: +49 (0)30 69 20 09 84 9

Internet: www.peritor.com E-Mail: [email protected]

Q&A