Where academic tradition
meets the exciting future

A Scalable Distributed M3 Platform on a Low-Power Cluster

Anders Berg, Petteri Karvinen, Stefan Grönroos, Frank Wickström, Natalia Díaz Rodríguez, Shohreh Hosseinzadeh, Johan Lilius, A Scalable Distributed M3 Platform on a Low-Power Cluster. In: Juha-Pekka Soininen Soininen, Sergey Balandin, Johan Lilius, Petri Liuha, Tullio Salmon Cinotti (Eds.), Open International M3 Semantic Interoperability Workshop , TUCS Proceedings 21, 49– 58, TUCS , 2013.

Abstract:

In order to facilitate the possibility for senior members of society to live at home, and lead an active life for as long as possible, it is often desirable to collect and process a large amount of sensor data. Sensor data includes vital signs such as heartbeat and blood pressure, as well as input from various in-home sensors to detect activities and events that might require further actions to be taken. In addition, high data-rate video streams might need to be stored and processed as well.
Most of this collected data could be stored and processed on-site without having to transport all data to a remote data center. In order to achieve this, we propose a compact data store consisting of one or several (depending on how much data storage and/or processing capacity is required)
low-power computing nodes consisting of hardware similar to that found in mobile phones. This choice of hardware allows us to create a small form-factor, easily hidden device with modest cooling requirements. Currently our design is based on the Smart-M3 software, extended to connect to a distributed RDF store backend. We are also re-implementing the subscription handler of Smart-M3 using the RETE algorithm to facilitate improved handling of a large number of subscriptions with a large amount of stored RDF triples. Furthermore, we are developing security and privacy models for the platform.
Figure 1 illustrates the intended high-level architecture of the proposed platform. This architecture spreads the data storage engine out over several nodes if needed, while the middleware and RETE engine are running on a master node. In future iterations, we hope to also distribute the RETE engine and most of the middleware over the cluster nodes. Thus far, we have implemented the connection of Smart-M3 to the 4store distributed RDF store, and finished most of the integration of the CLIPS RETE engine with Smart-M3. The implementation is run on a four-node cluster of HardKernel ODROID-X (quad-core ARM Cortex-A9 CPU) nodes. We are evaluating the performance of the clustered approach as well as identifying any performance bottlenecks that appear as a result of the choice of a low-power architecture compared to traditional server hardware.
A RESTful client API will be available to query the data store, as well as to set up subscriptions for interesting data. To increase interoperability with the applications developed within the Health and Wellbeing action line of the EIT ICT Labs, we also plan to connect the proposed platform to the Personal Health Labs (PHL) data store, in order to enable the exchange of data between the two platforms.

Files:

Full publication in PDF-format

BibTeX entry:

@INPROCEEDINGS{inpBeKaGrWiDxHoLi13a,
  title = {A Scalable Distributed M3 Platform on a Low-Power Cluster},
  booktitle = {Open International M3 Semantic Interoperability Workshop },
  author = {Berg, Anders and Karvinen, Petteri and Grönroos, Stefan and Wickström, Frank and Díaz Rodríguez, Natalia and Hosseinzadeh, Shohreh and Lilius, Johan},
  volume = {21},
  series = {TUCS Proceedings},
  editor = {Soininen, Juha-Pekka Soininen and Balandin, Sergey and Lilius, Johan and Liuha, Petri and Cinotti, Tullio Salmon},
  publisher = {TUCS },
  pages = {49– 58},
  year = {2013},
}

Belongs to TUCS Research Unit(s): Embedded Systems Laboratory (ESLAB)

Edit publication