HomeAncelus TopicsAncelus Benchmarks

Ancelus Benchmarks

  • Ancelus performance benchmarks are performed on a single server, single database, with randomized data and no pre-conditioning or sharding.

Published Benchmarks

  1. Throughput - Table Join Test
  2. Latency - single core, single thread
  3. Latency - multi-core, multi-thread
  4. Operational Data
  5. Relational Database Comparison
  6. Limits and Capacities

Throughput - Table Join Test:

This is the test that no other data management system will even try!  Industry analysts described it as "impressive", "amazing", "staggering."  Only Ancelus can do it.

Table Join Benchmarks

Intel Broadwell Server, dual socket, 44 cores, single disk channel. CentOS 7 operating system.

Total records: Table A - 1 million records, table B - 10 million, tables C,D,E,F,G - 1 billion records

Back to top

 Unit Operation Latency:

Single core, single thread, single field

Instead of using a "standardized" bundle of functions, Ancelus unit operations data should allow anyone to construct a performance estimate for any specific combination of transactions.

Unit Op Benchmarks

 (1) Scales as Log (n) with records.  Add ~500 nanoseconds for every 10X increase in records.

(2) Constant with size.

Intel Broadwell Server - two socket, 44 cores, single disk channel.  CentOS 7 operating system.

  Back to top

 Multi-Threaded Latency and Throughput (est)

Streaming throughput estimates are based on measured unit operation latency and average hardware scaling performance. Unit operations scale at approximately 94% for added cores deployed on a single CPU, 80% for added cores on second and subsequent CPUs.

 Multi core throughput

Assumes independent field writes (no contention for access to the same field)

Intel Broadwell Server - two socket, 44 cores, single disk channel. CentOS 7 operating system.

 Back to top

Operational Data

Administrative operations timing for setup, configure and load functions of the table join tests.

 Operational Data

1. Create, configure and activate all tables.

2. Populate all tables with randomly generated data.

3. Recover a database from the persistent store.

4. In replication mode, typically lass than 100 mSec.

Back to top

Comparison to Relational Database Performance:

Benchmark 3

 Back to top

Limits and Capacities

Current upper limit for a single server memory is about 3 TB, although the economic configuration is 1 TB.  In a relational database, assuming typical normalization, this would convert to about 5 to 10 TB.  For de-normalized big data systems the equivalent is about 20 TB.  Ancelus systems rarely require more than a single server for primary operations. 

 Limits  Capacities

Back to top

Go to top View Our Stats