improvement(performance): change c-s version to 3.17.15

[juliayakovlev]

Run throughput steps test with new 3.17.5 cassandra-stress version

Testing

• improvement(performance): change c-s version to 3.17.15 #10620 (comment)
• improvement(performance): change c-s version to 3.17.15 #10620 (comment)
• improvement(performance): change c-s version to 3.17.15 #10620 (comment)
• improvement(performance): change c-s version to 3.17.15 #10620 (comment)
• improvement(performance): change c-s version to 3.17.15 #10620 (comment)

PR pre-checks (self review)

• I added the relevant backport labels
• I didn't leave commented-out/debugging code

Reminders

• Add New configuration option and document them (in sdcm/sct_config.py)
• Add unit tests to cover my changes (under unit-test/ folder)
• Update the Readme/doc folder relevant to this change (if needed)

[juliayakovlev]

Scylla version 2025.2.0~dev-20250405, vnodes read load test.
Compare results with cassandra-stress versions 3.13.0 and 3.17.5.

Result:

• Slight latency reduction for 450K step
• 6% - 10% max read throughput improvement

C-s version	P99 read, 450K step	P99 read, 600K step	P99 read, 700K step	Max read throughput	Link
3.13.0	1.88	28.67	12205.42	794,618	link
3.17.5	1.30	33.85	42.83	873,679	link
3.17.5	1.07	33.44	39.68	839,075	link

[juliayakovlev]

Scylla version 2025.2.0~dev-20250405, tablets read load test.
Compare results with cassandra-stress versions 3.13.0 and 3.17.5.

Result:

• Slight latency reduction for 450K step
• 4.5% max read throughput improvement

C-s version	P99 read, 450K step	P99 read, 600K step	P99 read, 700K step	Max read throughput	Link
3.13.0	1.76	32.64	40.86	870,272	link
3.17.5	1.29	36.54	41.22	908,370	link

[juliayakovlev]

Scylla version 2025.2.0~dev-20250405, vnodes write load test.
Compare results with cassandra-stress versions 3.13.0 and 3.17.5.

Result:

• latency reduction for 200K & 300K steps. But the latency is too high, so I do not think it is
• 18% max write throughput improvement

C-s & java-driver version	P99 write, 200K step	P99 write, 300K step	Max read throughput	Link
2024.2.0~rc0-0.20240628.bdf707fe97c3 (java-driver: 3.11.5.3)	9571.40	992137.45	257,714	link
3.17.5 (java-driver: 3.11.5.7)	1718.62	258637.56	305,672	link

[juliayakovlev]

Scylla version 2025.2.0~dev-20250405, vnodes mixed load test.

Result:

• latency reduction for 200K & 300K steps. But the latency is too high, so I do not think it is
• 9% write/read max throughput reduction

C-s version	Max read throughput	Max write throughput	Link
3.13.0	305,547	305,526	link
3.17.5	277,007	277,045	link

[juliayakovlev]

Scylla version 2025.1.0~rc1-20250202, vnodes read load test.
I ran the test with 2025.1.0~rc1-20250202 where we received high latency for 450K step.

Latency for 450K step with cassandra-stress 3.17.5 is as expected.
But there is degradation about 8% in max read throughput

C-s version	P99 read, 450K step	P99 read, 600K step	P99 read, 700K step	Max read throughput	Link
3.13.0	20.45	40.86	42.70	925,061	https://argus.scylladb.com/tests/scylla-cluster-tests/a41ba05d-1146-45ed-a06d-387938983076
3.17.5	1.35	39.94	789.05	854,175	https://argus.scylladb.com/tests/scylla-cluster-tests/2d77ac47-a0f3-4d19-8825-422eb3a10dde