Ethernet over SpaceWire - Software Support
- Paper number
IAC-06-B5.7.02
- Author
Dr. Barry Cook, 4Links Limited, United Kingdom
- Coauthor
Mr. Paul Walker, 4Links Limited, United Kingdom
- Year
2006
- Abstract
Ethernet is a useful network mechanism that handles many well tried protocols, some of which are applicable to space. Re-using these protocols over SpaceWire could have major benefits in terms of cost, lead times, and risk. This paper will consider the software necessary to support networks that are physically SpaceWire while carrying Ethernet protocols. A companion paper will consider the hardware issues of running Ethernet over SpaceWire. These papers reflect the result of building an experimental network with SpaceWire physical layer and software support incorporated in a Linux device driver. Two major issues have been identified and will be addressed: routing packets and broadcast requests. To route packets, it is necessary to have a map of the network. We could assume that this will never change - once a spacecraft is launched it will not be altered - or that changes may occur due to failures or other reasons. There are good reasons to allow temporary or permanent changes for environmental or operational reasons; radiation effects or inclusion of cold-redundant units, for example. SpaceWire's grouped adaptive routing can help in some cases, but only provides limited opportunity for change. We prefer to allow the network to change and therefore to handle the changes dynamically. We implement a network discovery and route finding algorithm to map the current state of the network and allocate data flows. This allows both Plug-and-Play construction and operational network integrity checks with fault detection, isolation and recovery (FDIR). We will describe the simple algorithm that has been experimentally implemented and proven to achieve this. Broadcast can produce deadlock in networks with loops - such as multiply connected, fault tolerant SpaceWire networks. Many Ethernet protocols use broadcast and thus it must be supported. When we consider how often broadcast is actually used we conclude that it is a relatively rare event, such as emergency control or location discovery (ARP etc.). For such rare events using multiple unicast is an acceptable solution and we show how the above described mapping algorithm is used in this way to emulate broadcast. The authors have experience of building routing switches and test equipment for SpaceWire and Ethernet, and of building fault-tolerant dynamic plug-and-play systems for SpaceWire. From this experience, their conclusion is that it IS possible to run Ethernet over SpaceWire, and that it is possible to do so at reasonable cost.
- Abstract document
- Manuscript document
IAC-06-B5.7.02.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.