Reduce Your Cloud Spend with ScyllaDB by Tzach Livyatan

This talk will explore why ScyllaDB Cloud is a cost-effective alternative to DynamoDB, highlighting efficient design implementations like shared compute, local NVMe storage, and storage compression. It will also discuss new X Cloud features, better plans and pricing, and a direct cost comparison between ScyllaDB and DynamoDB

1. A ScyllaDB Community
Reduce Your Cloud
Spend with ScyllaDB
Tzach Livyatan
VP of Product

2. Agenda
■Reduce cost with ScyllaDB
■Reduce cost with ScyllaDB Cloud
■Coming Soon

3. Reduce cost with ScyllaDB

4. UNIQU
E
Guaranteed SLAPerfect horizontal & Vertical Scale
Query
Stream
Compac
tion
OLAP
task
CPU
Our Technology
1000 Nodes Cluster
8000 Cluster
K8S Deployment
60TB per Node 256 Cores per Node
1B Operations
per Second
Userspace
I/O Scheduler
Disk
Query
Commitlog
Compaction
Queue
Queue
Queue
0.5msec
Shards (per core)

5. Scylla Design Decisions
1C++ instead of Java

2All Things Async

3Shard per Core

4Unified Cache

5
I/O Scheduler

6
Autonomous

6. Scylla Design Decisions
1C++ instead of Java

2All Things Async

3Shard per Core

4Unified Cache

5
I/O Scheduler

6
Autonomous

7. Scylla Design Decisions
Threads







Shards
1C++ instead of Java

2All Things Async

3
4Unified Cache

5I/O Scheduler
6
Autonomous
Shard per Core

8. ScyllaDB Design Decisions
Legacy NoSQL Scylla
Key cache
Row cache
On-heap /
Off-heap
Linux page cache
SSTables
Unified cache
SSTables
Complex
Tuning
1
2All Things Async

3Shard per Core

4Unified Cache
5
I/O Scheduler

6
Autonomous
C++ instead of Java

9. Andy Pavlo
Are You Sure You Want to Use MMAP in Your Database Management System? (CIDR 2022)

10. ScyllaDB Design Decisions
Query
Commitlog
Compaction
Userspace
I/O
Scheduler
Disk
Max useful disk concurrency
I/O queued in
FS/device
No
queues
Queue
Queue
Queue
1
2All Things Async
3Shard per Core
4Unified Cache

5I/O Scheduler

6
Autonomous
C++ instead of Java

11. ScyllaDB is Different
11
■Multi queue
■Poll mode
■Userspace
■TCP/IP
■Thread per core
■lock-free
■Task scheduler
■Reactor programing
■C++14/17/20…
■NUMA friendly
■Log structured
allocator
■Zero copy
■DMA
■Log structured
■merge tree
■DB Aware cache
■Userspace I/O
■scheduler

12. Introducing
Tablets!

13. RAFT group
No. 299238
RAFT group
No. 299236 RAFT group
No. 299237
Tablets
key1key2
tablet
tablet
replica
tablet
replica

15. Replica Writes

16. Reduce cost with ScyllaDB Cloud

XCloud
Reduce Network Cost

17. XCloud Goals
■Reduce Costs with Just In Time scaling
■Reduce Costs with higher Storage utilization
■Improve UX with faster boot and scale

18. Mix Size + Fast Scaling =
Higher Storage Utilization

19. ScyllaDB Cloud
Available and Used Storage per zone
i4i.16xlarge i4i.16xlarge
i4i.16xlarge
i4i.16xlarge
i4i.16xlarge
i4i.16xlarge

20. ScyllaDB Cloud
Available and Used Storage per zone
Scale Trigger in 70%
Take a day
Initial reservation 15TB

21. ScyllaDB X Cloud
Available and Used Storage per zone
Scale Trigger in 90%
Effected Immediately

22. ScyllaDB X Cloud
Available and Used Storage per zone
i4i.16xlarge
I4i.2xlarge 1.875 TB
I4i.16xlarge 15 TB

23. ScyllaDB X Cloud
Available and Used Storage per zone
i4i.16xlarge
I4i.2xlarge 1.875 TB
I4i.2xlarge 1.875 TB

24. ScyllaDB X Cloud
Available and Used Storage per zone
i4i.16xlarge
I4i.2xlarge 1.875 TB
Consolidation to
I4i.4xlarge 3.75 TB

25. ScyllaDB X Cloud
Available and Used Storage per zone

26. ScyllaDB X Cloud
Allocated Resources

27. ScyllaDB X Cloud
XCloud Allocated Resources

28. ScyllaDB X Cloud

29. ScyllaDB X Cloud Summary
Tablets allow ScyllaDB Cluster to:
■Scale fast
■Run on heterogeneous servers

Which allows Scylla Cloud to:
■Just In Time Scaling
■Higher Storage Utilization

Which allow the user to Save Costs

30. Reduce Network Cost

31. Tablet File-based streaming
■Similar to Cassandra Zero-copy Streaming
■But better ;-)
■Tablets are always owned by the replica
■Simply copy, done.
■Up to x25 faster streaming compare to older versions

32. ScyllaDB X Cloud (2025.X)
Up to 3 times better compression, less network and storage usage.

33. Active/Active, replicated, auto-sharded
33
Multi-zone Deployment
us-east-1a
us-east-1b
us-east-1c

34. Active/Active, replicated, auto-sharded
34
Single-zone Deployment
us-east-1a

35. Coming Soon

Obj Storage Backend
Flexible Pricing

36. Today NVMe
■Local
■Ultra Fast
■Ephemeral
■Cheaper than EBS/SAN

Object Store / S3 based
■Remote
■Higher Latency
■Persistent
■Cheaper than SSD

NVMe vs S3 Obj Storage

37. S3 & Tablets = Win^2
tablet
tablet
replica
tablet
replica
Node
Node
Node
Node
Node
Node
Node

38. Tiered Storage
Cheap, Huge, Slow S3
■Latency < 100 ms
■No Size Limit
■Persistent, (more HA)
Expensive, Big, Fast SSD
■Latency < 1 ms
■Storage < 50TB per node
■Persistent, HA
Expensive, Small, Super Fast Cache
■Latency < 1 microsecond
■RAM < 1TB
■Not persistent

39. Scylla Cloud Pricing
$$$ On Demand > $$ Flex Credit > $ Reserved

40. Pricing On Demand
On Demand

41. Pricing Reserved
Reserved

42. Pricing Reserved + On Demand
Reserved
On Demand

43. Pricing Reserved + Flex
Reserved
Flex

44. Billing & Metering Progress

45. DynamoDB 50%
ScyllaDB Cloud guarantee to provide a 50% cost reduction compare to
DynamoDB

46. Summary
You can should use ScyllaDB to reduce cost in the cloud

ScyllaDB reduce costs using:
■Core Engineering
■Cloud Engineering
■Flexible Pricing

47. Thank You