Building a 300 node Raspberry Pi supercomputer; Robin Harris; In ZDNet; 2017-09-29.
Teaser: Commodity hardware makes possible massive 100,000 node clusters, because, after all, commodity hardware is “cheap” — if you’re Google. What if you want a lot of cycles but don’t have a few million dollars to spend? Think Raspberry Pi
Affordable and Energy-Efficient Cloud Computing Clusters: The Bolzano Raspberry Pi Cloud Cluster Experiment; Free University of Bozen-Bolzano, Bolzano, Italy; arXiv:1709.06815.
Pekka Abrahamsson, Sven Helmer, Nattakarn Phaphoom, Lorenzo Nicolodi, Nick Preda, Lorenzo Miori, Matteo Angriman, Juha Rikkilä, Xiaofeng Wang, Karim Hamily, and Sara Bugoloni.
- “Standard” 802.11 (wireline).
- Snowflake configuration.
A hierarchical star configuration.
- Consumer-grade 1Gb/s.
- Central meta-star switch
- Flash SDD is too slow.
- Must use NAS on HDD on the LAN.
- Custom PSU (not “stock” RPi PSU)
- Repurposed, used, higher-capacity PSUs.
- Subcluster: 24-nodes/PSU
- Count: 25 sub-clusters
Mounting (physical design)
- Think it through
- (stock) Debian v7
- Cannot run OpenStack
- Bespoke (bare metal) cluster management
- Some Paper; at Science Direct; no DOI, broken link.
Basit Qureshia, Yasir Javeda, Anis Koubàa, Mohamed-Foued Sritic, Maram Alajland; Performance of a Low Cost Hadoop Cluster for Image Analysis in Cloud Robotics Environment; In Proceedings of the Symposium on Data Mining Applications (SDMA2016); Riyadh, Saudi Arabia; 2016-03-30 (9 pages).
tl;dr → Claims to be able to run Hadoop and the Hadoop Image Processing Interface (HIPI) Library for Unmanned Aerial Vehicle (UAV) image processing.
- Ten (10) Amazing Raspberry Pi Clusters; Some Cub Reporter (SCR); In Network World; WHEN?
- Some Video; Hosted on YouTube; WHEN?
tl;dr → Something about using Legos for rack construction, for rack mounting; the physical design of the racks themselves.
- This is why you need to learn the Raspberry Pi 3
- An RPi 32-node Beowulf cluster
- An eight-node PS3 supercomputer