BIG DATA

PontusVision thread management improves system performance by up to 270% without any changes to hardware or software. This is a game changer in efficiency and latency for multi-threaded applications. It’s a patent pending, modeling tool that analyzes the hardware, optimizes the ways application threads are scheduled, and generates all the scripts to actually pin the threads to the right cores, which the OS does not do.

A FX trading platform in a tier 1 bank has seen a 270% reduction in latency, with ZERO code changes.

You can use for third party as well as in house codes. There’s no modification to the OS required, and the tool works with Windows, Linux, Solaris, and with any language (e.g. C/ C++ , Java, Perl, C#, even assembly).

The user inputs the software behaviour by using a drag-and drop GUI. The key inputs are communication between threads, and the CPU utilization of each thread.

The model attempts to keep applications that need to communicate with each other as close as possible, whilst avoiding context switches, and over-use of a core.

Also you would want to pin threads which are involved in I/O, (such as NIC or disc accesses), to be pinned to the CPU, (this is often only one of them), which is connected to the I/O bridge or hub.

All these parameters are used to create a latency score, which drives the model. Once the user is happy with the model results, the tool then outputs all the scripts required to actually tie the apps to the cores.

The technology is pretty trading centric, where it was developed, but this could be applicable to any multi-threaded application that runs on x86 hardware and is OS and language agnostic.

There are no kernel modules to worry about, infact no OS changes at all.