API library supports peer-to-peer comms to unify networks
Peer-to-peer data communication between multiple entities in a heterogenous computer system is supported by the open application programming interface (API) library developed by Ametek Abaco Systems. AXIS Takyon supports code development efficiency by unifying all network applications under a single API rather than multi-API solutions typically used today. This approach is claimed to maximise hardware availability in the system and reduce programming errors by reducing the amount of code needed. AXIS Takyon will be delivered as a new AXIS Pro software module.
The unified approach allows customers to do more in shorter time with less complexity, according to Ametek Abaco. The portable API allows developers to deploy Takyon as-is in different environments without major migration time, says the company. The AXIS Takyon API is intuitive in naming convention and concept and equipped with extensive documentation and example material, claimed to reduce average time to deployment from months to weeks. The Takyon API interface also simplifies the intercommunication of CPUs, GPUs and sub-systems to reduce the source code footprint and simplify the number of external APIs necessary to achieve the same functionality. This single, multi-purpose API replaces several APIs, says Ametek Abaco, to reduce the developer’s API toolbox and support faster deployment of peer-to-peer application software using the system’s hardware capabilities.
This product is designed to be one-way, zero-copy, and receive notification. It introduces predictable, measurable, and reproducible perturbation in send/receive activity. Fault tolerant hooks allow the API to detect timeouts and disconnects in node-to-node bonding. It can be used across a range of high-performance embedded computing (HPEC) applications including radar, sonar, signal integrity (SigInt) and image processing.
A fully open source reference implementation is available with an optimised, feature rich commercial implementation providing remote direct memory access (RDMA) support for low latency data movement between CPU and GPU compute entities.
