Transformer Refiner benchmarks

To help with capacity planning, Caplin have benchmarked Transformer Refiner in a range of use cases and scenarios.

Download the full report: Refiner 5.0 Benchmarks (PDF).

Caveat

While the benchmark tests were designed to emulate real-world traffic and user scenarios, they were conducted using specific hardware running in an isolated environment, and therefore no guarantees can be made that identical results will be achieved in other environments.

Benchmark hardware

Transformer Refiner was tested on a DELL PowerEdge R415 server with the following specification:

Transformer server hardware

Vendor

DELL

Model

PowerEdge R415

Processors

2 x AMD Opteron 4180 2.6GHz (hexa-core)

Memory

16GB

Operating System

CentOS 5.5 (Kernel 2.6.18-194.el5 64bit)

Results summary

The key findings are reproduced below. For the full results, see the report Refiner 5.0 Benchmarks (PDF).

  • For unique filter request rates of 8.3 per second, average response times on containers approaching the upper limit of 65,555 constituent objects remained below 100 milliseconds.

  • When filtering rapidly updating fields, performance deteriorates rapidly when Caplin Refiner has to process more than five million updates per second. The update rate is calculated by multiplying the number of unique filter requests by the number of underlying container objects that update per second.

  • When the Caplin Refiner module provides historic trade blotter data, two thousand users with one thousand blotter entries each produces two million unique objects, and requires about 16GB of random access memory. Although update latency remains low at these levels, available memory is the limiting factor.

  • Filtering and sorting on rapidly updating field values in large containers is not recommended.


See also: