Title: Building and evaluating a Raspberry Pi Cloud

Proposer: Hans-Wolfgang Loidl

Suggested supervisors: Hans-Wolfgang Loidl

Goal: To build a private cloud infrastructure out of networked Raspberry Pi2 nodes

Description:

The Raspberry Pi provides an interesting alternative to conventional hardware when building a private cloud hardware platform: it is cheap, widely available, and has a comparatively low energy consumption, while still providing considerable, though not high-end, compute power. It can therefore be used as a cheap in-house cloud platform for applications that require some cloud functionality but not top-end compute power. Several projects have already made use of this kind of setup (Glasgow Raspberry Pi Cloud).

The topic of this project is to build a local cloud infrastructure, based on Raspberry Pi2 devices, to use it as a parallel processing platform and to make systematic measurements of performance and of energy consumption. This will require systems knowledge in building a networked system, and parallel programming expertise in order to write and adapt parallel benchmarks. As a result, the thesis needs to identify strengths and weaknesses of this hardware platform for various application domains such as parallel program, cloud storage etc.

The project will proceed in the following stages:

• Review the use of Raspberry Pis and similar devices for building cloud infrastructures.
• Building the hardware infrastructure
• Installing a generic software infrastructure
• Designing and running experiments for performance and energy consumption
• Summarising the results from the experiments and drawing up conclusions on the usage of Raspberry Pi based private clouds.

Resources required: Networking hardware and power supplies for the infrastructure; RPis are available for loan in the department.

Degree of difficulty: Moderate

Background needed: General systems knowledge is required in configuring the networked system; for (parallel) performance measurements, simple benchmark programs should be developed, so this requires some working knowledge of parallel programming.

References: See hyperlinks above.