ScyllaDb vs Aerospike

ScyllaDB
1

I read the whitepaper about comparing ScyllaDb vs. Aerospike ScyllaDB White Paper | ScyllaDB vs. Aerospike: A NoSQL Database Performance Comparison, and it claims to overperform Aerospike from 30 to 40 percent. And I have some questions:

  1. Is it correct that Aerospike results were taken from the article Running Operational Workloads with Aerospike at Petabyte Scale in the Cloud on 20 Nodes August 2021 and weren’t running by Scylladb?
  2. What kind of versions were used for Scylladb and Aerospike?
  3. Where can I read about data type modeling for this comparison?
  4. What configurations were used for databases?
  5. Is it possible to reproduce this benchmark “locally”?
2
  1. Yes, Aerospike results were taken from the article Running Operational Workloads with Aerospike at Petabyte Scale in the Cloud on 20 Nodes August 2021 and they weren’t run by Scylladb.
  2. We used Scylla Enterprise version 2021.1.6, According to Aeorospike’s paper, the version they used is Aerospike Enterprise Server 5.4.0.2-1.
  3. The whitepaper (ScyllaDB White Paper | ScyllaDB vs. Aerospike: A NoSQL Database Performance Comparison) should have some high level description of the data model. It was intended to simulate an equivalent data model to what Aerospike described in their paper. It is not exactly the same workload as we used Cassandra stress tool to generate the workload.
  4. For ScyllaDB we used 20 x i3en.metal AWS instances for the database and 50 x c5n.9xlarge instances for the load generators (running cassandra stress). Aerospike report says they used 20 x i3en.24xlarge instances which are sized the same as i3en.metal.
  5. By locally, you mean on-premise? There should be no reason not to if you have strong enough servers and network. That said, I’d really recommend measuring a real-life workload that ressembles what you need rather than a synthetic benchmark.
3 Likes