← Back to article listing


March 7, 2023
Share this

The OnixS FIX Engine SDKs include reference implementation source code samples that are an aid for developers to quickly get familiar with, and then adapt to the target contexts.

 

One of those samples includes performance benchmarks.

 

We’re often asked, “what is your performance like?”

 

The best answer is to measure it yourself using the included reference implementation benchmark samples on the target platform(s).

 

That remains the recommendation, but these are some performance comparison numbers that we are commonly asked about so we’re sharing them here. 

 

 

Hardware

Intel(R) Core(TM) CPU i7-7700K @ 4.20GHz, 32 GB RAM

 

Software

Ubuntu 20.04 gcc 9.4.0

OnixS C++ FIX Engine x64 v 4.8.0
QuickFIX C++ x64 v 1.15.1

 

Throughput (msg/sec)

Message length: 131 bytes.

 

Mode 

OnixS 

QuickFIX

% OnixS faster

Send side

5.19964e+06

339183

15x

Receive side

3.45147e+06

258842

13x

 

Latency (microseconds)

Message length: 144 bytes.

 

Mode

OnixS

QuickFIX

% OnixS faster

Internal send min

0.10

0.58

5.8x

Internal send median

0.12

0.63

5.25x

Internal send 99%

0.13

0.74

5.6x

Overall send min

1.33

9.88

7.4x

Overall send median

1.45

10.20

7x

Overall send 99%

2.15

12.54

5.8x

RTT/2 min

1.08

4.08

3.77x

RTT/2 median

1.45

4.21

2.9x

RTT/2 99%

2.01

5.33

2.6x

 

 

Parsing (msg/sec)

 

Message size

OnixS

QuickFIX

% OnixS faster 

Small (106 bytes)

1e+07

1.11111e+06

9x

 

Notes:

  • There is no ability to measure internal receive latency in QuickFIX because there is no onReceivedBytes analog as in the OnixS FIX Engine.

  • The QuickFIX internal send latency, unlike the OnixS FIX Engine internal send latency, does not include the outgoing message serialization and session storage latencies because the toApp callback is called before these processes.

Is your FIX Engine specifically designed for ultra-low latency, high-frequency trading infrastructure?