2017 AWS DB Day | AWS 데이터베이스 개요 - 나의 업무에 적합한 데이터베이스는?

Databases onAWS
Blair Layton
Amazon Web Services, Business Development Manager
Database Services, Asia Pacific
Time : 10:10 – 11:00

Amazon RDS
Amazon DynamoDB Amazon Redshift
Amazon ElastiCache
Compute Storage
AWS Global Infrastructure
Database
Application Services
Deployment & Administration
Networking
AWS Database Services

If you host your databases on-premises
Power, HVAC, net
Rack & stack
Server maintenance
OS patches
DB s/w patches
Database backups
Scaling
High availability
DB s/w installs
OS installation
you
App optimization

If you host your databases in EC2
Power, HVAC, net
Rack & stack
Server maintenance
OS patches
DB s/w patches
Database backups
Scaling
High availability
DB s/w installs
OS installation
you
App optimization

If you choose a managed DB service like RDS
Power, HVAC, net
Rack & stack
Server maintenance
OS patches
DB s/w patches
Database backups
App optimization
High availability
DB s/w installs
OS installation
you
Scaling

differentiated effort increases the
uniqueness of an application

Relational Databases
Managed Service
Simple and fast to scale
Fast, predictable performance
Low cost, pay for what you use
Amazon
RDS
Amazon Aurora

Key Features
 Provision a database in 6 minutes
 Provision a MAZ database with a few mouse clicks
 Scale a database up/down with 60-90 seconds downtime
 Apply patches with 60-90 seconds downtime
 Add read replicas with a few mouse clicks
 Protect your backups and logs with 11 9’s of durability
 Recover to any point in time from nightly backups + logs
 Detailed metrics, down to 1 second intervals
 Secure your data with single click encryption at rest
Amazon
RDS

MySQL-compatible and PostgreSQL-compatible
relational database platform
Performance and availability of
commercial databases
Simplicity and cost-effectiveness of
open source databases
Delivered as a managed service
What is Amazon Aurora?

A service-oriented architecture applied to the database
Moved the logging and storage layer into a
multitenant, scale-out database-optimized
storage service
Integrated with other AWS services like
Amazon EC2, Amazon VPC, Amazon
DynamoDB, Amazon SWF, and Amazon
Route 53 for control plane operations
Integrated with Amazon S3 for continuous
backup with 99.999999999% durability
Control PlaneData Plane
Amazon
DynamoDB
Amazon SWF
Amazon Route
53
Logging + Storage
SQL
Transactions
Caching
Amazon S3
1
2
3

I/O traffic patterns: MySQL vs. Aurora
Binlog Data Double-write bufferLog records FRM files, metadata
T Y P E O F W R I T E S
EBS mirrorEBS mirror
AZ 1 AZ 2
Amazon S3
MYSQL WITH STANDBY
SEQUENTIAL
WRITE
SEQUENTIAL
WRITE
EBS
Amazon Elastic Block
Store (EBS)
Primary
Instance
Standby
Instance
AZ 1 AZ 3
Primary
Instance
Amazon S3
AZ 2
Replica
Instance
AMAZON AURORA
ASYNC
4/6 QUORUM
DISTRIBUTED WRITES

Key Features
 Highly available storage, with 6 copies across 3 AZs
 No more MySQL replica lag!
 Auto-scaling storage for IOPS and capacity
 Parallel, distributed & asynchronous recovery
 Survivable caches: quick restart & no brown outs
 Faster, more predictable failover
 Up to 5x faster than MySQL and more than 2x faster than PostgreSQL on
the same hardware
 Cheaper at scale!

Recent Announcements
 Performance enhancements: Fast DDL, fast index build, spatial
indexing, hot row contention
 Availability features: Zero-downtime patching, database cloning (Q2),
database backtrack (Q2)
 Eco-system integration: Load from S3, IAM integration (Q2), select into
S3 (Q2), log upload to CloudWatch Logs & S3 (Q2)
 Cost reduction: t2.small – cuts cost of entry by half – you can run Aurora
for $1 / day
 Growing footprint: London, Montreal, Ohio, and San Francisco – now
available in all 3AZ regions

2/3 of top 100 AWS customers
8 of top 10 gaming customers
Fastest growing service in AWS history

Feature Aurora MySQL MariaDB PostgreSQL Oracle SQL Server
VPC      
High availability      
Instance Scaling      
Encryption      
Read replicas    
Oracle Golden
Gate / DMS
DMSCross region
replicas
   
Max Storage 64 TB 6 TB 6 TB 6 TB 6 TB 4 TB
Scale Storage Auto scaling    
Provisioned IOPS NA 30,000 30,000 30,000 30,000 20,000
Largest Instance R3.8XL R3.8XL
M4.10XL
R3.8XL
M4.10XL
R3.8XL
M4.10XL
R3.8XL
M4.10XL
R3.8XL
M4.10XL
RDS Engine Capability Matrix

 Automates sampling of data
 Exposes data via API
 Provides UI to show Database Load
Database Load:
Performance Insights at a glance

2017 AWS DB Day | AWS 데이터베이스 개요 - 나의 업무에 적합한 데이터베이스는?

In-Memory Cache
Elastic and reliable
Memcached or Redis
Managed Service
Amazon
ElastiCache

Why In-Memory Caching?
Everything is connected - phones, tablets, cars, air
conditioners, toasters
Demand for real-time performance – online games, ad tech,
eCommerce, social apps, etc.
Load is spikey and unpredictable
Database performance is often the bottleneck

Memcached vs. Redis
• In-memory
• Multithreaded / Multicore
• No persistence
• String values
• Easy horizontal clustering
• In-memory
• Single-threaded
• Read replicas
• Clustering
• Persistence
• Atomic operations
• Advanced data types -
http://redis.io/topics/data-types
• Pub/sub messaging

RDS and ElastiCache are Behind Grab’s Taxi-Booking App
The latency of a cab call must be
low, and remain low even in times of
peak traffic of hundreds of
thousands of cab requests per
minute. We use ElastiCache for
Redis in front of RDS MySQL to keep
our systems’ real time performance
at any scale.
Ryan Ooi
Sr. Devops Engineer, Grab
”
“ • Grab is a popular taxi hailing app in southeast Asia.
• Average response time of the API layer is <40ms, mandating an
in-memory layer to achieve such performance.
• A small devops team that tried running Redis on EC2 before, but
that was too much work. Using both RDS and ElastiCache in
Multi-AZ allowed them to outsource all the management to AWS.

NoSQL Database
Durable low latency
Managed Service
Massive and seamless scalability
Amazon
DynamoDB

=
• Consistent, single-digit millisecond
latency at any scale
• Highly durable and available—3x
replication
• Accessible via simple and powerful
APIs
• Supports both document and key-
value data models
• No table size or throughout limits

Consistent low latency whether
scaling up/down or operating at your
provisioned limits
Durable Low Latency – At Scale

Relational data warehouse
Massively parallel; petabyte scale
Managed Service
HDD and SSD platforms
$1,000/TB/year; starts at $0.25/hour
Amazon
Redshift
a lot faster
a lot cheaper
a whole lot simpler

Key Capabilities
 Scaleable: Petabyte scale data warehouse: from 160 GB to 2PB
 Fast: parallel execution with compressed, sorted data on optimized
hardware
 Inexpensive: Start from $0.25/hour or $1,000/TB/year
 Managed Service: Easy to provision, backup, restore, patch and scale
 Secure: Load and store encrypted data, SSL, Audit logging
 Innovative: 100 new features since launch!
 Large Ecosystem: Major data integration and visualization ISVs support
Redshift with a large consulting partner base

Amazon.com – Weblog analysis
Web log analysis for Amazon.com
1PB+ workload, 2TB/day, growing 67% YoY
Largest table: 400 TB
Want to understand customer
behavior
Solution
Legacy DW—query across 1 week/hr.
Hadoop—query across 1 month/hr.

Query 15 months of data (1PB) in 14 minutes
Load 5B rows in 10 minutes
21B rows joined with 10B rows – 3 days (Hive) to 2 hours
Load pipeline: 90 hours (Oracle) to 8 hours
64 clusters
800 total nodes
13PB provisioned storage
2 DBAs
Data warehouses
can be
fast
and
simple

Databases on EC2
 Any database that runs on Windows or Linux!
 Many AMIs available from technology partners
 Oracle Database, MS SQL Server, MongoDB, Vertica, …
 White papers available on best practices
 Oracle Database, MS SQL Server, MongoDB, Cassandra, …
 Why?
 No managed service
 Full control
 Exceed limits of managed service, e.g. > 6TB of storage on RDS

New X1 Instance - Tons of Memory
• Designed for large-scale, in-memory
applications in the cloud
• Ideal for in-memory databases like SAP HANA
and big data processing apps like Spark and
Presto
• Powered by Intel® Xeon® E7 8880 v3 Haswell
processors
• Features up to 2TB of memory and up to 128
vCPUs per instance.
• Expanding to 4TB in 2017. With 8TB and 16TB
in the works too!

AWS Database Migration Service (AWS DMS)
DMS migrates databases to AWS easily and
securely with minimal downtime. It can migrate
your data to and from most widely used
commercial and open-source databases.
Amazon Aurora Amazon DynamoDB

Less than $10 per TB
with CDC!

AWS Schema Conversion Tool (AWS SCT)
SCT helps automate many database schema and
code conversion tasks when migrating between
database engines or data warehouse engines
Amazon Aurora

SCT helps with converting tables, views, and code
Sequences
User Defined Types
Synonyms
Packages
Stored Procedures
Functions
Triggers
Schemas
Tables
Indexes
Views

SCT can tell you how hard the migration will be
1. Connect SCT to
Source and Target
databases.
2. Run Assessment
Report.
3. Read Executive
Summary.
4. Follow detailed
instructions.

Which Service Should You Use?
Situation Solution
Existing application
Use your existing engine on RDS
• MySQL  Amazon Aurora, RDS for MySQL
• PostgreSQL  RDS for PostgreSQL
• Oracle, SQL Server  RDS for Oracle, RDS for SQL Server
New application
• If you can avoid relational features  DynamoDB
• If you need relational features  Amazon Aurora
Data Warehouse & BI • Amazon Redshift and Amazon QuickSight
Ad hoc analysis of data in S3 • Amazon Athena and Amazon QuickSight
Spark, Hadoop, Hive, HBase • Amazon EMR
Log analytics, operational
monitoring and search
• Amazon Elasticsearch Service

추천서비스 고군분투기
onAWS
Jinwoo Park
RecoBell

반도체 쪽으로 유학을 가려던 평범한 전기공학도
어쩌다한 창업
지금은 레코벨에서 빅데이터를 경험중
AWS 사용 6년째
Me?

• 추천서비스
• 개인화 마케팅 서비스
• 검색광고 솔루션
• Retargeting 광고
• 대용량 푸시 서버
++ 데이터를 이용한 돈벌이
데이터가 돈이 된다는 것을 증명하는 사람들

0
50
100
150
200
250
300
350
2014. 12 2015. 3 2015. 9 2016. 4 2016. 10 2017. 5

추천서비스 사용 사이트 숫자
0
50
100
150
200
250
300
350
2014. 12 2015. 3 2015. 9 2016. 4 2016. 10 2017. 5

추천서비스 고군분투기 on AWS
이 세션에서 다룰 내용
IDC 에서 클라우드로
최초 10개 미만이던 사이트 숫자를 300개 이상으로
빠르게 성장한 이야기

데이터 분석은 크게 3 단계
Logging Analysis
Operation
Visualize

Legacy on IDC
Server on IDC
Logger
API
DB
(+Engine) Client Site
User Behavior
Result (HTML)
JS SDK
• 대기업위주의 SI
• Engineer 보다는 Data Scientists 위주
• 간간히 알바/외주를 써서 몇 개의
고객사에 대해서만 IDC를 빌려서 구현

Legacy on IDC
Server on IDC
Logger
API
DB
(+Engine) Client Site
User Behavior
Result (HTML)
JS SDK
• 고객사 영업되면, 세팅 한 세월
• 장애 대응 한 세월
• 테스트 한 세월
• 내가 SE 도 아니고..

AWS 로 이사가기
Server on IDC
Logger
API
DB
(+Engine)
• AWS 로 이사가기로 결정
• Logger / API 는 묶어서 사용
• DB 는 RDS 를 이용

• 관리용이성
• 뛰어난 확장성
• 가용성 및 내구성
• 빠른속도(SSD)
• 보안
• 저렴한 비용
Amazon RDS

AWS 위 첫 번째 집
• 큰 규모의 업체는 SI
• 작은 규모의 업체는
클라우드
MySQLs
Logger API

하지만 가정은 보통 깨어지고..
MAU 천만 업체 등장

AWS 위 두 번째 집
• 서비스 쪽과 엔진 쪽 분리
• LogDB + ResultDB
• Rec. Engine
MySQLs
Logger API
Partial Log
Recommendation Result
Recommendation Engine

• 빠른속도(SSD)
• 저렴한 비용 => SSD $0.138/GB.Month
• + 긴 RDS 생성시간
Amazon RDS

대대적인 구조 개선 시작

구조개선 목표
• Logger : 안정성 및 TPS
• Engine : 속도 및 안정성
• API : Business 로직 최소화

• 간편성
• 내구성
• 확장가능
• 보안
• 가용성
• 저렴한비용
• 간편한 데이터 전송
• 통합
• 손쉬운관리
S3 (Simple Storage Service)

AWS 위 세 번째 집
• S3 에 데이터 저장
Infra on AWS
Logger API
Client Site
User Behavior Result (JSON)
EngineLog Bucket Result Bucket

Logger
Kinesis
PUT
Producer Consumer
S3
Log Bucket

Real-time Streaming Data Buffer
• 실시간성
• 사용편의성
• 병렬처리
• 탄력성
• 저렴한비용
• 안정성
Amazon Kinesis Stream

Engine
• EC2 : c4.8xlarge
– 36 CPU / 132 ECU / 60G Mem.
– Provisioned IOPS EBS (3000)
• PostgreSQL + Spring Boot
PUT
S3
Result Bucket
GET
S3
Log Bucket
…/{siteId}/{recType}/{item or userId}.json…/{siteId}/{logType}/{date}/{time}.json

API
GET
S3
Result Bucket
…/{siteId}/{recType}/{item or userId}.json
API
"recType": "a002",
"itemId": "354427372",
"results”:
[
{
"itemId": "362054013",
"categoryId": "5502167”,
"rank": 1,
"score": 1,
http://{apiUrl}/{recType}?itemId={itemId}Req
Res

성공적으로 20개 사이트 라이브
MAU 천만이 넘어도 문제 없음
이때만든로거는지금까지도문제없음

상황 1
• 가장 쉽게 이 상황에 대처하는 방법?
– Logger / API 는 Scale-out
– Engine 220개를 띄운다. => 먼저 EC2 instance 의 Limit 을 풀어야한다.
• c4.8xlarge instance : $2.3/hour
– AWS 과금형식이 시간당 과금
– 200 개 사이트 중에 수 분 내에 엔진이 끝나는 사이트도..
– + Provisioned IOPS 도 비싸

상황 2
• 추천 고객사가 다양해 지면서 다양한 e-Commerce 고객사 등장
• e-Commerce 가 아닌 고객사도 등장
• Data Schema 가 다들 다름
• 이 때 레코벨의 개발자 2명
• Java + SpringBoot + PostgreSQL 수정이 너무 어려워

두마리 토끼
• 더 가격이 싼 구조
– 비싼 c4.8xlarge 대체
– Provisioned IOPS 도 대체
• 개발자 없이, 빠르게 고객사의 요구사항을 들어줄 수 있는 구조
– Data Scientist 들이 SQL 은 빠르게 수정할 수 있으니,
최대한 SQL 로 모든걸 해결할 수 있게하자.

Data Warehouse 서비스
• 신속함
• 저렴함
• 간편성
• 탄력성
• 보안
• 호환성 (PostgreSQL)
Amazon Redshift

AWS 위 네 번째 집
• 1/6 가격
• 쉬운 수정
• 개발자 2명
• DS 4명
• 사이트 220개EngineLogger Log bucket Result Bucket API
추천데이터
로그

• 간편성
• 내구성
• 확장가능
• 보안
• 가용성
• 저렴한비용
• 간편한 데이터 전송
• 통합
• 손쉬운관리
S3 (Simple Storage Service)
• 100만개의 상품을 가진 사이트면, 한번 엔진이 돌때마다 비용이..
– $0.0047 / 1,000 * 1,000,000 = $4.7
• 하루에 세번, 한달이면..
– $4.7 * 3 * 30 = $423

고성능, 고-확장성 DB
• 고성능
• 뛰어난 보안
• MySQL/PostgreSQL 호환
• 뛰어난 확장성
• 높은 가용성 및 내구성
• 완전 관리형
RDS : Amazon Aurora

AWS 위 다섯 번째 집
EngineLogger Log bucket Result
On Aurora
API
추천데이터
로그

Remind & Tips
• RDS / EC2 의 IOPS 에 항상 주의
• 웬만한 데이터는 S3, 하지만 PUT 가격에 주의
• 데이터 버퍼로써 Kinesis 활용
• Aurora 는 요금체계가 기존 RDS 와 다르다. (IOPS 과금)
• Redshift 는 쉽고 빠르다.

Thank you!
개발자2명
하루약100기가의로그
당신을기다리고있습니다.

2017 AWS DB Day | AWS 데이터베이스 개요 - 나의 업무에 적합한 데이터베이스는?

Recommended

More Related Content

More from Amazon Web Services Korea (20)

Recently uploaded (20)

2017 AWS DB Day | AWS 데이터베이스 개요 - 나의 업무에 적합한 데이터베이스는?