Attend this session for a technical deep dive about RDS Postgres and Aurora Postgres. Come hear from Mark Porter, the General Manager of Aurora PostgreSQL and RDS at AWS, as he covers service specific use cases and applications within the AWS worldwide public sector community. Learn More: https://aws.amazon.com/government-education/

1. © 2016, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Mark Porter, General Manager
Amazon Aurora, Amazon RDS
June 14, 2017
Managed PostgreSQL in the Amazon Cloud
Amazon RDS for PostgreSQL & Amazon Aurora
PostgreSQL

2. RDS – Amazon Relational Database Service
We make it easy to set up, operate, and scale relational databases in the cloud.
We automate the time-consuming database administration tasks so you can focus
on application optimization. We (try to) do it with a higher bar of security, durability,
and availability than you can achieve either on-premises or on EC2.
You choose the database that is right for you: Amazon Aurora, MySQL, MariaDB,
Oracle, Microsoft SQL Server, and PostgreSQL. We give you choices from tiny
databases on small CPUs to enterprise-class databases on enterprise hardware.
We work to save you money through elasticity, operational efficiencies, and a deep
passion for delighting you.

3. If you run your databases on-premises
Power, HVAC, net
Rack & stack
Server maintenance
OS patches
DB s/w patches
Database backups
High availability
DB s/w installs
OS installation
you
Scaling
Hardware
management
Operating system
management
Database
management
App optimization
Database deep dive
Design consultation
Best practicesWhat you
want to
spend your
time on…

4. Power, HVAC, net
Rack & stack
Server maintenance
OS patches
DB s/w patches
Database backups
Scaling
High availability
DB s/w installs
OS installation
If you run your databases on-premises
?
you
Hardware
management
Operating System
management
Database
management
App optimization
Database deep dive
Design consultation
Best practicesWhat you
want to
spend your
time on…

5. And then you move to the cloud…
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
Database
management
Your cloud
vendor
App optimization
Database deep dive
Design consultation
Best practicesWhat you
want to
spend your
time on…
Hardware
management
Operating system
management

6. OS patches
DB s/w patches
Database backups
Scaling
High availability
DB s/w installs
App optimization
Power, HVAC, net
Rack & stack
Server maintenance
OS installation
And then you move to the cloud…
? Hardware
management
Operating system
managementDatabase
management
you
Your cloud
vendor
Database deep dive
Design consultation
Best practicesWhat you
want to
spend your
time on…

7. And now move to RDS…
Power, HVAC, net
Rack & stack
Server maintenance
OS patches
DB s/w patches
Database backups
High availability
DB s/w installs
OS installation
Scaling
Database deep dive
Design consultation
App optimization
Best practices
Hardware
management
Operating system
management
Database
management
What you
want to
spend your
time on…
you
Your cloud
vendor

8. A single management platform for all your databases
 Single
Pane of
Glass
 Console
or API
 Monitor
via email,
SMS

9. Manageable
 Rapid deployment with pre-configured parameters
 Patch management
 Monitoring and metrics
Available and durable
Scalability
Secure
 Encryption in transit and at rest
 TDE with Oracle Database and SQL Server
Inexpensive
Elastic
Easy to script
RDS key features

10. DB Snapshots
 User-driven snapshots of database
 Kept until explicitly deleted
Automated Backups
 Nightly system snapshots + transaction
backup
 Enables point-in-time restore to any point in
retention period, up to the last 5 minutes
 Max retention period = 35 days
Backups and recovery

11. Enterprise-grade fault tolerance solution for production databases
With a few clicks, Amazon RDS creates and synchronously
maintains a standby in a different Availability Zone
Automatic failover (~60-90 sec) in case of:
 Loss of availability in primary AZ
 Loss of connectivity to primary
 Host or storage failure on primary
 Vertical scaling of CPU or storage
 Software patching
High availability: Multi-AZ deployments

12. Scale nodes vertically up or down
 t2.small (1 virtual core, 2GiB)
 m3.2xlarge (8 virtual cores, 30GiB)
 r3.8xlarge (32 virtual cores, 244GiB)
 r4.16xlarge (64 virtual cores, 488GiB)
 1.32xlarge (~128 virtual cores, ~1952 GiB)
• (Coming soon)
Convert storage to PIOPS or GP2
 Consistent throughput + low I/O latencies
Scale storage vertically with minimal downtime
 Increase throughput by spreading data across additional
volumes
 Independently scale provisioned IOPS
Push-button scaling

13. Horizontal scaling with Read Replicas
Add Read Replicas with a single button push
• Horizontal scaling of read-heavy workloads
• Offload reporting
• Cross Region as well (VPN, etc.)
Available for MySQL, MariaDB, PostgreSQL,
Aurora
Overcoming challenges
• RDS makes it easy to re-create if fallen behind
• Deploy a proxy to round robin requests
• With Aurora, promote chosen read replica to
master

14. But RDS is only as good as the engines it
manages….

15. Amazon RDS
RDS Platform
OPEN SOURCE ENGINES COMMERCIAL ENGINES
• Advanced monitoring
• Routine maintenance
• Push-button scaling
• Automatic failover
• Backup & recovery
• X-region replication
• Isolation & security
• Industry compliance
• Automated patching
CLOUD NATIVE ENGINE

16. Agenda
 Why did we build Amazon Aurora?
 Why add PostgreSQL compatibility?
 Durability and availability architecture
 Performance results vs. PostgreSQL
 Performance architecture
 Announcing Performance Insights
 Feature Roadmap
 Questions
+

17. Traditional relational databases are hard to scale
Multiple layers of
functionality all in a
monolithic stack
SQL
Transactions
Caching
Logging
Storage

18. Traditional approaches to scale databases
Each architecture is limited by the monolithic mindset
SQL
Transactions
Caching
Logging
SQL
Transactions
Caching
Logging
Application Application
SQL
Transactions
Caching
Logging
SQL
Transactions
Caching
Logging
Storage
Application
Storage Storage
SQL
Transactions
Caching
Logging
Storage
SQL
Transactions
Caching
Logging
Storage

19. Reimagining the relational database
What if you were inventing the database today?
You would break apart the stack
You would build something that:
 Can scale out…
 Is self-healing…
 Leverages distributed services…

20. A service-oriented architecture applied to the database
Move the logging and storage layer into a
multitenant, scale-out, database-optimized
storage service
Integrate with other AWS services like
Amazon EC2, Amazon VPC, Amazon
DynamoDB, Amazon SWF, and Amazon
Route 53 for control & monitoring
Make it a managed service — using Amazon
RDS. Takes care of management and
administrative functions.
Amazon
DynamoDB
Amazon SWF
Amazon Route 53
Logging + Storage
SQL
Transactions
Caching
Amazon S3
1
2
3
Amazon RDS

21. In 2014, we launched Amazon Aurora with MySQL compatibility.
Now, we are adding PostgreSQL compatibility.
Customers can now choose how to use the Amazon
cloud-optimized relational database, with the performance and
availability of commercial databases and the simplicity and cost-
effectiveness of open source databases.
Making Amazon Aurora better

22. Start with the customer – Why add PostgreSQL?

23. Start with the customer – Why add PostgreSQL?

24. Open source database
In active development for 20 years
Owned by a foundation, not a single company
Permissive innovation-friendly open source license
PostgreSQL fast facts
Open Source Initiative

25. High performance out of the box
Object-oriented and ANSI-SQL:2008 compatible
Most geospatial features of any open source database
Supports stored procedures in 12 languages (Java, Perl,
Python, Ruby, Tcl, C/C++, its own Oracle-like PL/pgSQL,
etc.)
PostgreSQL fast facts

26. Most Oracle-compatible open source database
Highest AWS Schema Conversion Tool automatic
conversion rates are from Oracle to PostgreSQL
PostgreSQL fast facts

27. What does Aurora PostgreSQL mean? (1 of 2)
PostgreSQL 9.6 + Amazon Aurora cloud-optimized storage
Performance: Up to 2x+ better performance than PostgreSQL alone
Availability: Failover time of < 30 seconds
Durability: 6 copies across 3 Availability Zones
Read Replicas: Single-digit millisecond lag times on up to 15 replicas
Fully compatible with PostgreSQL, now and for the foreseeable future
Not a compatibility layer — native PostgreSQL implementation
Amazon Aurora Storage

28. What does Aurora PostgreSQL mean? (2 of 2)
Cloud-native security and encryption
AWS Key Management Service (KMS) and AWS
Identity and Access Management (IAM)
Easy to manage with Amazon RDS
Easy to load and unload
AWS Database Migration Service and AWS Schema
Conversion Tool
Fully compatible with PostgreSQL, now and for the
foreseeable future
Not a compatibility layer — native PostgreSQL
implementation
AWS DMS
Amazon RDS
PostgreSQL

29. Amazon Aurora PostgreSQL
Durability & availability

30. Scale-out, distributed, log structured storage
Master Replica Replica Replica
Availability Zone 1
Shared Storage Volume – Transaction Aware
Primary
Database
Node
Read
Replica /
Secondary
Node
Read
Replica /
Secondary
Node
Read
Replica /
Secondary
Node
Availability Zone 2 Availability Zone 3
AWS Region
Storage
Monitoring
Database and
Instance
Monitoring

31. Amazon Aurora Storage Engine overview
Data is replicated 6 times across 3 Availability
Zones
Continuous backup to Amazon S3
(built for 11 9s durability)
Continuous monitoring of nodes and disks for
repair
10 GB segments as unit of repair or hotspot
rebalance
Quorum system for read/write; latency tolerant
Quorum membership changes do not stall writes
Storage volume automatically grows up to 64 TB
AZ 1 AZ 2 AZ 3
Amazon S3
Database
Node
Storage
Node
Storage
Node
Storage
Node
Storage
Node
Storage
Node
Storage
Node
Storage
Monitoring

32. What can fail?
Segment failures (disks)
Node failures (machines)
AZ failures (network or data center)
Optimizations
4 out of 6 write quorum
3 out of 6 read quorum
Peer-to-peer replication for repairs
SQL
Transaction
AZ 1 AZ 2 AZ 3
Caching
SQL
Transaction
AZ 1 AZ 2 AZ 3
Caching
Amazon Aurora Storage Engine fault tolerance

33. Amazon Aurora Replicas
Availability
Failing database nodes are automatically
detected and replaced
Failing database processes are
automatically detected and recycled
Replicas are automatically promoted to
primary if needed (failover)
Customer specifiable failover order
AZ 1 AZ 3AZ 2
Primary
Node
Primary
Node
Primary
Database
Node
Primary
Node
Primary
Node
Read
Replica
Primary
Node
Primary
Node
Read
Replica
Database
and
Instance
Monitoring
Performance
Customer applications can scale out read traffic
across read replicas
Read balancing across read replicas

34. Amazon Aurora Continuous Backup
Segment snapshot Log records
Recovery point
Segment 1
Segment 2
Segment 3
Time
• Take periodic snapshot of each segment in parallel; stream the logs to Amazon S3
• Backup happens continuously without performance or availability impact
• At restore, retrieve the appropriate segment snapshots and log streams to storage nodes
• Apply log streams to segment snapshots in parallel and asynchronously

35. Faster, more predictable failover with Amazon Aurora
App
RunningFailure Detection DNS Propagation
Recovery
Database
Failure
Amazon RDS for PostgreSQL is good: failover times of ~60 seconds
Replica-Aware App Running
Failure Detection DNS Propagation
Recovery
Database
Failure
Amazon Aurora is better: failover times < 30 seconds
1 5 - 2 0 s e c 3 - 1 0 s e c
App
Running

36. Amazon Aurora PostgreSQL
Performance

37. PostgreSQL
Benchmark system configurations
Amazon Aurora
AZ 1
EBS EBS EBS
45,000 total IOPS
AZ 1 AZ 2 AZ 3
Amazon S3
m4.16xlarge
database
instance
Storage
Node
Storage
Node
Storage
Node
Storage
Node
Storage
Node
Storage
Node
c4.8xlarge
client driver
m4.16xlarge
database
instance
c4.8xlarge
client driver
ext4 filesystem
m4.16xlarge (64 VCPU, 256GiB), c4.8xlarge (36 VCPU, 60GiB)

38. Amazon Aurora is >=2x faster on PgBench
pgbench “tpcb-like” workload, scale 2000 (30GiB). All configurations run for 60 minutes

39. Amazon Aurora is 2x-3x faster on SysBench
Amazon Aurora delivers 2x the absolute peak of PostgreSQL and 3x
PostgreSQL performance at high client counts
SysBench oltp (write-only) workload with 30 GB database with 250 tables and 400,000 initial rows per table

40. Amazon Aurora: Over 120,000 writes/sec
OLTP test statistics:
queries performed:
read: 0
write: 432772903
other:(begin + commit) 216366749
total: 649139652
transactions: 108163671 (30044.73 per sec.)
read/write requests: 432772903 (120211.75 per sec.)
other operations: 216366749 (60100.40 per sec.)
ignored errors: 39407 (10.95 per sec.)
reconnects: 0 (0.00 per sec.)
sysbench write-only 10GB workload with 250 tables and 25,000 initial rows per table. 10-minute warmup, 3,076 clients
Ignored errors are key constraint errors, designed into sysbench
Sustained sysbench throughput over 120K writes/sec

41. Amazon Aurora loads data 3x faster
Database initialization is three times faster than PostgreSQL using the
standard PgBench benchmark
Command: pgbench -i -s 2000 –F 90

42. Amazon Aurora gives >2x faster response times
Response time under heavy write load >2x faster than PostgreSQL
(and >10x more consistent)
SysBench oltp(write-only) 23GiB workload with 250 tables and 300,000 initial rows per table. 10-minute warmup.

43. Amazon Aurora has more consistent throughput
While running at load, performance is more than three times
more consistent than PostgreSQL
PgBench “tpcb-like” workload at scale 2000. Amazon Aurora was run with 1280 clients. PostgreSQL was run with
512 clients (the concurrency at which it delivered the best overall throughput)

44. Amazon Aurora is 3x faster at large scale
Scales from 1.5x to 3x faster as database grows from 10 GiB to 100 GiB
SysBench oltp(write-only) – 10 GiB with 250 tables & 150,000 rows and 100 GiB with 250 tables & 1,500,000 rows
75,666
27,491
112,390
82,714
0
20,000
40,000
60,000
80,000
100,000
120,000
10GB 100GB
writes/sec
SysBench Test Size
SysBench write-only
PostgreSQL Amazon Aurora

45. Amazon Aurora delivers up to 85x faster recovery
SysBench oltp (write-only) 10GiB workload with 250 tables & 150,000 rows
Writes per Second 69,620
Writes per Second 32,765
Writes per Second 16,075
Writes per Second 92,415
Recovery Time (seconds) 102.0
Recovery Time (seconds) 52.0
Recovery Time (seconds) 13.0
Recovery Time (seconds) 1.2
0 20 40 60 80 100 120 140
0 20,000 40,000 60,000 80,000
PostgreSQL
12.5GB
Checkpoint
PostgreSQL
8.3GB
Checkpoint
PostgreSQL
2.1GB
Checkpoint
Amazon Aurora
No Checkpoints
Recovery Time in Seconds
Writes Per Second
Crash Recovery Time - SysBench 10GB Write Workload
Transaction-aware storage system recovers almost instantly

46. Amazon Aurora with PostgreSQL compatibility
Performance by the numbers
Measurement Result
PgBench >= 2x faster
SysBench 2x-3x faster
Data Loading 3x faster
Response Time >2x faster
Throughput Jitter >3x more consistent
Throughput at Scale 3x faster
Recovery Speed Up to 85x faster

47. Amazon Aurora PostgreSQL
Performance architecture

48. Do fewer I/Os
Minimize network packets
Offload the database engine
DO LESS WORK
Process asynchronously
Reduce latency path
Use lock-free data structures
Batch operations together
BE MORE EFFICIENT
How does Amazon Aurora achieve high performance?
DATABASES ARE ALL ABOUT I/O
NETWORK-ATTACHED STORAGE IS ALL ABOUT PACKETS/SECOND
HIGH-THROUGHPUT PROCESSING NEEDS CPU AND MEMORY OPTIMIZATIONS

49. Write I/O traffic in Amazon RDS for PostgreSQL
WAL DATA COMMIT LOG & FILES
RDS FOR POSTGRESQL WITH MULTI-AZ
EBS mirrorEBS mirror
AZ 1 AZ 2
Amazon S3
EBS
Amazon Elastic
Block Store (EBS)
Primary
Database
Node
Standby
Database
Node
1
2
3
4
5
Issue write to Amazon EBS, EBS issues to mirror,
acknowledge when both done
Stage write to standby instance
Issue write to EBS on standby instance
I/O FLOW
Steps 1, 3, 5 are sequential and synchronous
This amplifies both latency and jitter
Many types of writes for each user operation
OBSERVATIONS
T Y P E O F W R IT E

50. Write I/O traffic in an Amazon Aurora database node
AZ 1 AZ 3
Primary
Database
Node
Amazon S3
AZ 2
Read
Replica /
Secondary
Node
AMAZON AURORA
ASYNC
4/6 QUORUM
DISTRIBUTED
WRITES
DATAAMAZON AURORA + WAL LOG COMMIT LOG & FILES
I/O FLOW
Only write WAL records; all steps asynchronous
No data block writes (checkpoint, cache replacement)
6X more log writes, but 9X less network traffic
Tolerant of network and storage outlier latency
OBSERVATIONS
2x or better PostgreSQL Community Edition performance on
write-only or mixed read-write workloads
PERFORMANCE
Boxcar log records – fully ordered by LSN
Shuffle to appropriate segments – partially ordered
Boxcar to storage nodes and issue writes
WAL
T Y P E O F W R IT E
Read
Replica /
Secondary
Node

51. Asynchronous group commits
Read
Write
Commit
Read
Read
T1
Commit (T1)
Commit (T2)
Commit (T3)
LSN 10
LSN 12
LSN 22
LSN 50
LSN 30
LSN 34
LSN 41
LSN 47
LSN 20
LSN 49
Commit (T4)
Commit (T5)
Commit (T6)
Commit (T7)
Commit (T8)
LSN GROWTH
Durable LSN at head-node
COMMIT QUEUE
Pending commits in LSN order
TIME
GROUP
COMMIT
TRANSACTIONS
Read
Write
Commit
Read
Read
T1
Read
Write
Commit
Read
Read
Tn
TRADITIONAL APPROACH AMAZON AURORA
Maintain a buffer of log records to write out to disk
Issue write when buffer full or time out waiting for writes
First writer has latency penalty when write rate is low
Request I/O with first write, fill buffer till write picked up
Individual write durable when 4 of 6 storage nodes ACK
Advance DB Durable point up to earliest pending ACK

52. Write I/O traffic in an Amazon Aurora storage node
LOG RECORDS
Primary
Database
Node
INCOMING QUEUE
STORAGE NODE
AMAZON S3 BACKUP
1
2
3
4
5
6
7
8
UPDATE
QUEUE
ACK
HOT
LOG
DATA
BLOCKS
POINT IN TIME
SNAPSHOT
GC
SCRUB
COALESCE
SORT
GROUP
PEER TO PEER GOSSIPPeer
Storage
Nodes
All steps are asynchronous
Only steps 1 and 2 are in foreground latency path
Input queue is far smaller than PostgreSQL
Favors latency-sensitive operations
Uses disk space to buffer against spikes in activity
OBSERVATIONS
I/O FLOW
① Receive record and add to in-memory queue
② Persist record and acknowledge
③ Organize records and identify gaps in log
④ Gossip with peers to fill in holes
⑤ Coalesce log records into new data block versions
⑥ Periodically stage log and new block versions to Amazon
S3
⑦ Periodically garbage collect old versions
⑧ Periodically validate CRC codes on blocks

53. I/O traffic in Aurora Replicas
PAGE CACHE
UPDATE
Aurora Master
30% Read
70% Write
Aurora Replica
100% New Reads
Shared Multi-AZ Storage
PostgreSQL Master
30% Read
70% Write
PostgreSQL Replica
30% New Reads
70% Write
SINGLE-THREADED
WAL APPLY
Data Volume Data Volume
Physical: Ship redo (WAL) to Replica
Write workload similar on both instances
Independent storage
Physical: Ship redo (WAL) from Master to Replica
Replica shares storage. No writes performed
Cached pages have redo applied
Advance read view when all commits seen
POSTGRESQL READ SCALING AMAZON AURORA READ SCALING

54. Applications restart faster with survivable caches
Cache normally lives inside the
operating system database process–
and goes away when/if that database
dies
Aurora moves the cache out of the
database process
Cache remains warm in the event of a
database restart
Lets the database resume fully loaded
operations much faster
Cache lives outside the database
process and remains warm across
database restarts
SQL
Transactions
Caching
SQL
Transactions
Caching
SQL
Transactions
Caching
RUNNING CRASH AND RESTART RUNNING

55. Amazon Aurora PostgreSQL
Announcing Performance Insights

56. First step: Enhanced Monitoring
Released 2016
O/S Metrics
Process & thread list
Up to 1 second granularity

57. Second step: Performance Insights
Database Engine
Performance Tuning

61. Beyond database load
• Lock detection
• Execution plans
• API access
• Included with RDS
• 35 days data retention
• Support for all RDS database engines in 2017

62. Amazon Aurora PostgreSQL
Roadmap

63. High
Performance
Easy
to Operate &
Compatible
High
Availability
Secure
by Design
Amazon Aurora with PostgreSQL compatibility —
Launch Roadmap
 2-4x faster than PostgreSQL on large
instances
 Up to 64 TB of storage per instance
 Write jitter reduction
 Near synchronous replicas (<25ms)
 Reader endpoint
 Enhanced OS monitoring
 Performance Insights
 Push-button migration
 Auto-scaling storage
 Continuous backup and PITR
 Easy provisioning / patching
 All PostgreSQL features
 All RDS for PostgreSQL extensions
 AWS DMS supported inbound
 Failover in less than 30 seconds
 Customer specifiable failover order
 Up to 15 readable failover targets
 Instant crash recovery
 Survivable buffer cache
 X-region snapshot copy
 Encryption at rest (AWS KMS)
 Encryption in transit (SSL)
 Amazon VPC by default
 Row level security

64. Amazon
Aurora
Available
Durable
The Amazon Aurora database family
AWS DMS Amazon RDS
AWS IAM,
KMS, & VPC
Amazon
S3
Convenient
Compatible
Automatic
Failover
Read
Replicas
X
6 Copies
High
Performance
& Scale
Secure
Encryption at rest
and in transit
Enterprise
Performance
64TB
Storage
PostgreSQL
MySQL

65. Thank You!
Mark Porter
markpor@amazon.com