Skip to main content
A ScyllaDB Community
ScyllaDB Engineering:
What's New, What's Next
Yaniv Kaul
VP R&D
Yaniv Kaul
■ ScyllaDB’s VP R&D in the last 3 years
■ Previously, managed the development of multiple
storage, virtualization and cloud products at Red Hat
as well as lead the cross engineering performance
group.
■ New features of ScyllaDB 2025.x
■ A glimpse into MORE new features of ScyllaDB 2025.x
■ New features of ScyllaDB Cloud
■ A glimpse at 2026 and beyond
Presentation Agenda
Tablets - the new data layout standard
Vector Search
Disk and Network Compression
Raft-based features
New features of ScyllaDB 2025 releases
TABLETS
Tablets - a real game changer
It’s not very often that we change a core architecture!
Tablets are a completely new way of distributing data.
Tablets provide:
■ Unprecedented elasticity
■ Ability to use the disks up to 90% utilization
■ In 2026: workload based resource balancing,
incremental repair and more
TCO
Tablets - what are they?
■ A new way of keeping the token range mapping for
a set of replicas of a table’s data.
■ Stored in system.tablets
■ Configured per keyspace, with additional per table configuration options
■ Tablets go through split and merge, in order to keep their sizes in range
■ Compaction, repair, backup tasks are all tablets aware
■ All ScyllaDB drivers are tablets aware. Backwards compatible with others
■ Streaming is file based - orders of magnitude faster, hardly consumes CPU
■ Cleanup is now a simple file deletion
25%
Table A
Table B
Table C
40%
35%
Tablets for amazingly fast elasticity
■ Parallel bootstrap/scaling of nodes
■ Parallel decommission in 2026.1
■ File-based streaming is (too*) fast!
■ No cleanup of un-owned needed!
■ Support different instance sizes in the same cluster
■ No resharding needed!
■ CQL and Alternator access is available right away
■ We still use heat-weighted-load-balancing
* it’s so fast, we limit it to 75% of base instance network speed
Scaling a cluster
now takes
minutes,
not hours or days!
Tablets provide improved disk utilization
■ Thanks to fast elasticity and size based load
balancing (2026.1 feature), we can keep disk
utilization to maximum.
■ In ScyllaDB X Cloud, we allow up 90% disk utilization
BEFORE scaling out the cluster
■ New nodes join the cluster in minutes
■ Scaling with small instance size increments
■ Availability zones scale individually
■ Out of space prevention for edge cases
Vector
Search
Vector Search as part of ScyllaDB Cloud
■ AI use cases:
■ Feature store
■ Time series
■ Vector Search (similarity)
■ Generally available in ScyllaDB
Cloud with 2025.4
■ Filtering, quantization and
instance resizing - 2026.1
TCO
4 Simple steps
■ Create the table - with vector
■ Insert some data - with embedding
■ Create the vector index
■ Query!
CREATE TABLE IF NOT EXISTS myapp.comments (
record_id timeuuid,
id uuid,
commenter text,
comment text,
comment_vector vector<float, 64>,
created_at timestamp,
PRIMARY KEY (id, created_at)
);
INSERT INTO myapp.comments (
record_id,
id,
commenter,
comment,
comment_vector,
created_at
) VALUES (
now(),
uuid(),
'Alice',
'I like vector search in ScyllaDB.',
[0.12, 0.34, 0.56, 0.78, 0.91, 0.15, 0.62, 0.48,
0.22, 0.31, 0.40, 0.67, 0.53, 0.84, 0.19, 0.72,
0.63, 0.54, 0.26, 0.33, 0.11, 0.09, 0.27, 0.41,
0.69, 0.82, 0.57, 0.38, 0.71, 0.46, 0.55, 0.64,
0.17, 0.81, 0.23, 0.95, 0.66, 0.35, 0.44, 0.59,
0.02, 0.75, 0.28, 0.16, 0.92, 0.88, 0.47, 0.13,
0.99, 0.21, 0.32, 0.83, 0.45, 0.04, 0.86, 0.25,
0.36, 0.73, 0.07, 0.61, 0.52, 0.14, 0.68, 0.05],
toTimestamp(now())
);
CREATE CUSTOM INDEX IF NOT EXISTS ann_idx
ON myapp.comments(comment_vector)
USING 'vector_index'
WITH OPTIONS = { 'similarity_function': 'DOT_PRODUCT' };
SELECT id, commenter, comment, created_at
FROM myapp.comments
ORDER BY comment_vector ANN OF [ 0.11, 0.35, 0.56, 0.77, … ]
LIMIT 5;
STEP
04
STEP
03
STEP
02
STEP
01
Embedding
Vector
Index
Advanced Compression
TCO
Node to node network compression
■ ZSTD or LZ4 - with dictionary
■ Configurable
■ Limited CPU usage
■ Not within rack (AZ)
■ Not for small packets
Data at rest (sstable) compression
■ Default is now LZ4 + dictionary
■ ~20% better than LZ4
■ ZSTD + dictionary is even better!
■ Especially for JSON / text based data
■ Online conversion
Dictionary based advanced compression
lZ4 49%
LZ4
w/dict
44%
ZSTD 34%
ZSTD
w/dict
29%
Raft-based strongly
consistent features
With Raft, the following are now strongly consistent
■ Schema changes
■ Topology changes (gossip going away in 2026.2!)
■ Authentication (no need to change RF for auth tables manually!)
■ Service levels
■ Tablets location
■ More to come in 2026…
Native object storage backup/restore
New cloud instances and machine support
Encryption with cloud provider keys
Drivers
More new features of ScyllaDB Enterprise
Native object storage (S3) client within ScyllaDB
C++ based, per shard, part of the streaming I/O scheduling group client:
■ Co-operates, does not compete, with the rest of the workloads
■ FAST - can easily saturate* the network interfaces
■ Supports AWS S3, Google GCS (2026.1), more in the future.
■ ScyllaDB Manager 3.6+ uses it automatically for backup
■ Backup now completes x5 times faster, with negligible impact to the workload
* it’s so fast, we limit it to 75% of base instance network speed
Modern instance types support
■ AWS I8g, i8ge - provide 60-70% higher throughput than i4i/i3en
■ AWS I7i, i7ie support - improved disk performance (with 4K blocks)
■ GCP Z3
■ Azure LSv4
■ Oracle image and performance tweaks - 2026.1
■ 512 cores per node - 2026.1
Encryption at rest (EaR) improvements
■ Encryption at rest can now use AWS, GCP or Azure provided keys for
encrypting SSTables.
■ ScyllaDB Cloud now encrypts all new clusters by default - either with
ScyllaDB KMS keys or the customer’s.
New Drivers:
■ Rust Driver fully GA
■ Our fastest driver!
■ C++ over Rust
■ Builds on the same Rust code
■ Replaces legacy C++ driver
with full compatibility
■ To be followed with Node JS
over Rust, Python over Rust
■ CDC Rust library
Existing drivers improvements:
■ Native Tablets support
■ Vector type support
■ Performance improvements
■ Removal of legacy protocols
■ Compression to Alternator
requests and responses
(2026.1)
Drivers
X Cloud
Tablets + new instances + Raft based topology + new scaling algorithm
+ flexible billing + parallel boot + fast startup + automation =
Amazing elasticity
Pick the size - 2 steps only
1. Minimum vCPU count
2. Minimum storage size
That’s it.
Scaling in action
Billing:
metering & contract mapping
Vector Search
■ Part of Standard and X Cloud
■ Multiple VS node size options
■ Resize if needed
■ 1 node per Availability Zone
Billing and Metering
■ Contract mapping
■ Flex credits usage
■ Metering improvements
API
■ API support for all operations
■ ScyllaDB Terraform provider to
ease API consumption and
operations
■ Better BYOA integration
■ Bring Your Own Key (BYOK)
More ScyllaDB Cloud improvements
A glimpse into 2026
■ Vnodes to Tablets online migration
■ STRONG consistency
■ Amazingly fast K-V store
■ More features to reduce TCO, across the board
■ example - tiering to lower cost storage
■ Workload isolation:
■ Isolate hot data from other data, by utilization tablets load balancing
■ Workload based scaling in ScyllaDB Cloud
■ More Real Time AI integrations and features
In the works - major themes
Stay in Touch
Yaniv Kaul
yaniv.kaul@scylladb.com
@YanivKaul
https://github.com/mykaul
https://www.linkedin.com/in/ykaul/
Your Big Slide Title
Goes Here
Your Slide Title
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Vestibulum
dictum ex leo, ac blandit arcu convallis et. Donec faucibus porttitor lorem
vitae luctus.
■ Vestibulum ante ipsum primis in
■ Faucibus orci luctus et ultrices posuere
■ Cubilia curae Nulla facilisi
■ Urna volutpat
■ Vel suscipit lorem condimentum
Your Title Goes Here
Your Big Slide Title
Goes Here
Your Big Slide Title
Goes Here
Your Big Slide Title
Goes Here
Your Big Slide Title
Goes Here
Your Big Slide Title
Goes Here
ScyllaDB Monsters
ScyllaDB Summit 2025 Styles
The default body font is Roboto Condensed.
You can adjust the size as needed.
You can also use Roboto (the uncondensed version).
For code you should use Roboto Mono and you can set it on
this dark background
Header font is Archivo:
Monster Scale Color palette (base for the gradient): Accent colors:
#15D2A8 #0D72FF #7933FF #0F1040 #F0581E #CE1576
#A3E3FF
https://fonts.google.com/specimen/Archivo
#FFB32B
Body copy font is Roboto
ScyllaDB Logo
Stay in Touch
Your First and Last Name
Your Email Address
Your Twitter Handle
Your GitHub Handle
Your LinkedIn URL
Your photo
goes here,
smile :)
Your First and Last Name
Your Email Address
Your Twitter Handle
Your GitHub Handle
Your LinkedIn URL
Your photo
goes here,
smile :)