12–16 Jun 2016
Gothenburg, Sweden
Europe/Amsterdam timezone

A Lightweight Operating System for the SSDP

16 Jun 2016, 09:35
30m
Gothenburg, Sweden

Gothenburg, Sweden

DSP Day: DSP software, tools and libraries Session 5: DSP Software and Applications

Speaker

Mr Armin Luntzer (University of Vienna)

Description

A common problem of space missions is the limited processing power of available space-qualified hardware. In recent years, ESA has been pursuing the development of an ITAR-free next generation payload processor. One of the outputs of this effort is a prototype SoC called the MPPB (Massively Parallel Processor Breadboard) developed by Recore System and built around the Very Long Instruction Word Xentium DSP architecture designed by the same company. In this platform, a LEON processor is acting as a supervisor, controlling a Network-on-Chip (NoC) with multiple DSPs, memory and I/O devices attached to it. In the course of the NGAPP (Next Generation Astronomy Processing Platform) activities, an evaluation of the MPPB was performed in a joint effort of RUAG Space Austria and the Department of Astrophysics at the University of Vienna. It was found that, given the highly innovative nature of this new processing platform, a novel approach was needed regarding the management of system resources, DMA mechanics and DSP program design for best efficiency and turnover rates. Consequently, the University of Vienna developed an experimental operating system to stably drive the DSP cores and the MPPB close to its performance limit. This is achieved by splitting processing tasks into a pipeline of small units (kernels) that are dynamically scheduled to run on the Xentium DSPs as required by the amount of data in the pipeline stages, thereby overcoming bottlenecks resulting from memory transfer overheads and cache sizes that would inevitably emerge when using large, monolithic programs with the particular characteristics of the MPPB. At present, activities are carried out by Thales Alenia Space España and Recore Systems in an effort to create the Scalable Sensor Data Processor (SSDP), where an ASIC with adapted specifications is being developed from the MPPB. In order to support this new hardware, a more refined version of the experimental operating system is under development at the University of Vienna that aims to become fully space-qualifiable, supporting applicable documentation and S/W standards. The software will be tailored to the NoC concept present in the SSDP and is to be optimised for best performance in key areas of system and resource management. These include fast and efficient interrupt handling to ensure fast response times and high memory throughput for DMA transfers that service the Xentium data caches and fast I/O interfaces like SpaceWire or ADC/DAC. Supporting functionality, for example device drivers, threads and schedulers, timing and a system configuration/information interface will be provided. Effort will be made to keep CPU and memory footprints at a minimum, so the LEON processor is available for duties other than DSP and data processing control, such as handling of telecommands or instrument related control tasks. A major aim is to make the operating system as easy to use as possible by providing appropriate, well designed interfaces in order to keep the need for configuration and extra programming effort at a minimum. To encourage use, modification and redistribution of the operating system, it will be made available under an open-source license, including all drivers, modules and available DSP kernels.

Summary

The Department of Astrophysics at the University of Vienna is a provider
of payload instrument flight software. Among the projects under development is
a custom, lightweight operating system for the upcoming Scalable Sensor Data
Processor (SSDP) based on prior experience with its predecessor, the MPPB.
The objective of this project is to create easy to use software that is capable
of efficiently driving the SSDP's Xentium DSP cores.

Primary author

Mr Armin Luntzer (University of Vienna)

Co-authors

Dr Christian Reimers (University of Vienna) Prof. Franz Kerschbaum (University of Vienna) Dr Roland Ottensamer (University of Vienna)

Presentation materials