The Mechanics of Scale by Dominik Tornow

Scaling distributed systems introduces massive complexity. This talk builds a clear mental model of concurrency, key challenges like consistency vs. availability, and essential techniques like partitioning and replication—helping you design and scale systems with confidence.

1. A ScyllaDB
Community
The Mechanics of Scale
Dominik Tornow
Founder & CEO

2. Correctness
Safety Liveness

3. Distributed Systems
The What

4. C6C5C4C3C2C1C

5. C1S1 C2 S2
C3S3 C4 S4
C5S5 C6 S6
Network

6. Distributed Systems
The Why

7. Correctness
ScalabilityReliability

8. ! "
ConcurrencyPartial OrderCoordination
DistributionPartial FailureRecovery

9. Concurrency
P | Q

10. Distribution
P | Q | ⚡

11. P = a • b • c
Q = x • y • z

12. a • b • c • x • y • z
a • b • x • c • y • z
a • b • x • y • c • z
a • b • x • y • z • c
a • x • b • c • y • z
a • x • b • y • c • z
a • x • b • y • z • c
a • x • y • b • c • z
a • x • y • b • z • c

13. a
x
a • b
a • x
x • a
x • y
a • b • c
a • b • x
a • x • b

14. Recv
Req
Step w/
Side Effect
Send
Res
Send
Req
Recv
Res

15. Recv
Req
Step w/
Side Effect
Send
Res
Send
Req
Rec
Res

16. Recv
Req
Step w/
Side Effect
Send
Res
Send
Req
Rec
Res

17. Recv
Req
Step w/
Side Effect
Send
Res
Send
Req
Rec
Res

18. Recv
Req
Step w/
Side Effect
Send
Res
Send
Req
Rec
Res

19. Recv
Req
Step w/
Side Effect
Send
Res
Send
Req
Rec
Res
?

20. Scalability & Reliability
From First Principles

21. Redundancy
DuplicationCoordination

22. Logic
Gate
y = f(x1, x2, ... xn)

23. Logic
Gate
Logic
Gate
Logic
Gate
Logic
Gate
Majority
Gate

24. Logic
Gate
Logic
Gate
Logic
Gate
Logic
Gate
Majority
Gate
Majority
Gate
Majority
Gate

25. z = 3x = 1y = 2
x = 1
y = 2
z = 3

26. x = 1
y = 2
z = 3
x = 1
y = 2
z = 3
x = 1
y = 2
z = 3

27. x = 1x = 1x = 1
y = 2y = 2y = 2
z = 3z = 3z = 3
PartitioningPartitioningPartitioning
Replication
Replication
Replication

28. Client
DataDataData
ServerServerServer
ClientClient
Partitioning distributes the load
Replication distributes the data

29. Partitioning

30. Logical
Object
Physical
Object1
Physical
Object2
Physical
Objectn
One logical object is represented by multiple physical objects
Physical Object1 + Physical Object2 + Physical Objectn = Logical Object

31. Replication

32. Logical
Object
Physical
Object1
Physical
Object2
Physical
Objectn
One logical object is represented by multiple physical objects
Physical Object1 = Physical Object2 = Physical Objectn = Logical Object

33. Conclusion

34. Stay in Touch
Dominik Tornow
[email protected]
@DominikTornow
dtornow
https://www.linkedin.com/in/dtornow/