Evan Tschannen

�1

• Worked on FoundationDB for 8 years

• Touched every core component of the database

• Recently added multi-region asynchronous replication

Evan Tschannen

• https://www.foundationdb.org

• https://github.com/apple/foundationdb

• Open Source, Apache License (v2)

Why are there so many databases?

!4

!5

• Tension between performance, scalability, and consistency

• A wide variety of data models

Why are there so many databases?

FoundationDB Performance

• Organizes many instances of a single process database into a single database

• Read and write throughput 90% of the aggregate individual read and write throughput

• Single hop read latencies

• Four hop write latencies

FoundationDB is a…

• Highly performant,

• Scalable,

• Ordered key value store,

• That supports ACID transactions.

!8

A

B

C

Writer

Servers

Reader

Wait for 2 of 3 responses Wait for 2 of 3 responses

!9

A

B

C

Writer

Servers

Reader

Wait for 2 of 3 responses Wait for 2 of 3 responses

!10

A

B

C

Writer

Servers

Reader

Wait for 3 of 3 responses Wait for 1 of 3 responses

!11

A

B

C

Writer

Servers

Reader

Wait for 3 of 3 responses Wait for 1 of 3 responses

!12

A

B

C

Writer

Servers

Reader

Wait for 3 of 3 responses Wait for 1 of 3 responses

Versions(Stored in a different database)

D

0: A,B,C 101: B,C,D

!13

Coordinators

Storage Servers

ZooKeeper (Failure Handling)

SQLite (Reads)

Transaction Logs

FoundationDB (Writes)

!14

Coordinators

Storage Servers Transaction Logs

Cluster Controller

!15

Coordinators

Storage Servers Transaction Logs

Cluster Controller

!16

Get Read Version

ReadWrite

Commit

FoundationDB APIVersions

200

400

100 A : apple(200)

(A, 200)

(B, banana)

!17

Get Read Version

ReadWrite

Commit

FoundationDB APIVersions

200

400

100 A : apple(200)

(A, 200)

(B, banana)300 A : orange

!18

Coordinators

Storage Servers

Cluster Controller

Master

Proxies Resolvers Transaction Logs

!19

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

!20

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

(200)

(A, 200)

(B : banana)

Data Distribution

!21

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

(200)

(A, 200)

(400)

(B : banana)

!22

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

(200)

(A, 200)

(400)

(B : banana)

!23

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

(200)

(A, 200)

(B : banana)

!24

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

(200)

(A, 200)

(400)

(B : banana)

!25

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

!26

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

(400)

(300)

(200)

!27

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

(400)

(400)

(300)

(200)

!28

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

!29

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

!30

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

!31

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

Data Distribution

!32

Coordinators

Cluster Controller

Master

Proxies Resolvers Transaction LogsStorage Servers

Get Read Version

Reads

Writes

Commit

FoundationDB Performance

• Organizes many instances of a single process database into a single database

• Read and write throughput 90% of the aggregate individual read and write throughput

• Single hop read latencies

• Four hop write latencies

Does it work?

• Run the entire distributed database in a single process

• Inject random failures

• Check correctness

• Reproduce errors

https://www.foundationdb.org

!35