Performance analysis of single board computer clusters

Basford, Philip J.; Johnston, Steven J.; Perkins, Colin; Garnock-Jones, Tony; Tso, Fung Po; Pezaros, Dimitrios P.; Mullins, Robert; Yoneki, Eiko; Singer, Jeremy; Cox, Simon J.

doi:10.1016/j.future.2019.07.040

Cited by 55 publications

(41 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…[10] However, in some cases, SBC is used for the computing section; to reduce computation power volume for a portable cluster. [5] For example, the single board can also use for clustering or test application similar to supercomputing techniques alike Open Multi-Processing (OpenMP) and Message Passing Interface (MPI). [11] This research will implement the SBC testing operation and configuration as follow.…”

Section: Methodsmentioning

confidence: 99%

“…The study presented a Pi Stack, which reduces the computing power and low cost to support edge computing applications, and it also has individual heartbeat monitoring in Pi Stack. [5] HPL is a benchmark that runs to find solutions to systems of linear equations = when A is a dense matrix or matrix where most of its members are not 0, with the time complexity of working as O(n 3 ) when using HPL to run on the device will get one FLOPS number. HPL is considered a suitable measure method since most scientific work presently implied a lot of linear system equations and measurement by floating-point numbers.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Patternless Piezoelectric Energy Harvester for Ultra Low Frequency Applications

Awal

Jusoh

Sabapathy

et al. 2020

IJIE

View full text Add to dashboard Cite

This paper presents a pattern less piezoelectric harvester for ultra low power energy applications. Usually patterned cantilevers are used as vibration energy harvester which results additional fabrication process. Hence, to reduce the process, a four layer cantilever configuration is used to design the harvester with Aluminum, Silicon and Zinc Oxide. The device dimension is settled to 12×10×≈0.5009 mm3 with ≈300 nm deposition thickness for each layer. The modeling and fabrication processes are demonstrated in detail. The induced voltage by the cantilever is obtained through the analytical and practical measurements. From the measurements, it is found that, the maximum induced voltage is 91.2 mV from practical measurement with voltage density of 1.517 mV/mm3. It is evident from the results that, this pattern less model can be useful for next generation vibration energy harvester with simpler technology.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A Patternless Piezoelectric Energy Harvester for Ultra Low Frequency Applications

Awal

Jusoh

Sabapathy

et al. 2020

IJIE

View full text Add to dashboard Cite

show abstract

“…For the Internet case, M0 device configuration was used for simulation on all nodes, while for IoT devices three types of configurations M1-M3 were applied. Device parameters were selected on the basis of the real hardware used in Industrial Edge Computing studies [21], [22] and Cyber-Physical systems [23], [24] used for studying connected automated vehicles, together with large scaled video analysis platforms; their values are presented in Table I. Device M1 represents Raspberry PI 3 hardware, M2 is a LinkIt Smart 7688 Du IoT device, and M3 is ESP32 SoC. These values represent the maximum values of the parameters the device can reach.…”

Section: B Controllable Parametersmentioning

confidence: 99%

Performance Analysis of Edge Computing in IoT

Skirelis¹,

Navakauskas²

2020

ELEKTRON ELEKTROTECH

View full text Add to dashboard Cite

Upcoming 5G technology and the demand for near real-time response services raise the need for optimizing current IoT solutions. The aim of this paper is to model and execute the performance analysis of network structures suitable for Edge Computing in IoT. The prior research into different topology and parameter sets have not provided sufficient clarity, on which parameters had a considerable impact on overall system performance; therefore, repetitive simulations were performed to express dispersion of alternating values, as well as determining its confidence intervals. The paper presents Edge Computing service simulation setup on known and newly derived network topologies with different complexity varying network bandwidth and network delay parameters. The experimental investigation has revealed that the IoT configuration network is more sensitive to network topology, while the Internet configuration is more sensitive to network parameters. The discussion on the results received debates possible causes of performance differences in network parameters and device configurations, the comparison to similar state-of-the-art research results has also been presented. Finally, conclusions with recommendations based on the results acquired have been provided.

show abstract

“…The algorithm is also suitable to run on smaller processors. The Raspberry Pi is capable of

0.132 G F L O P s

[43] and the ARM‐core Cortex‐A9 (a common mobile phone processor) is capable of

0.372 G F L O P s

[44]; therefore, in this example, the optimisation algorithm uses 11.9% and 4.2% of the respective processors' capacity. Note that although this optimisation algorithm is implementable in real time, it still requires a large amount of computation compared to other calculations undertaken to update the control signal: for example, calculating

u (t)

for a given value of

Θ (t)

and

x (t)

, through (51) and (7), requires only 334 floating point operations per time‐step (

3.34 \times 10^{- 6} G F L O P s

).…”

Section: Control Designmentioning

confidence: 99%

Robust fault tolerant control allocation for a modern over‐actuated commercial aircraft

Vile

Alwi

Edwards

2020

IET Control Theory & Appl

View full text Add to dashboard Cite

This paper presents a novel form of control allocation, designed within a sliding mode framework, for the fault tolerant control of over‐actuated systems. The control allocation is designed in such a way as to allow a subset of the actuators to remain inactive under nominal fault‐free conditions. In the event that the active set of actuators becomes unable to provide the desired performance, an adaption process takes place which allows the inactive actuators to compensate. A computationally light gradient descent algorithm is proposed to govern the adaption which guarantees that, if possible, actuator saturation is avoided and system performance is maintained, even in the event of severe actuator faults/failures. Rigorous conditions are derived, in terms of the faults/failures, uncertainties in fault reconstruction information and the adaptive process, which ensure sliding occurs in a finite time and that the resulting motion is stable. To demonstrate the effectiveness of the control scheme, a high‐fidelity blended wing body aircraft model is also proposed in this paper; this particular configuration of aircraft is nominally unstable, with poor control authority and a large amount of redundancy, making it a suitable candidate for testing reconfigurable fault tolerant control laws in the presence of input constraints.

show abstract

Performance analysis of single board computer clusters

Cited by 55 publications

References 6 publications

A Patternless Piezoelectric Energy Harvester for Ultra Low Frequency Applications

A Patternless Piezoelectric Energy Harvester for Ultra Low Frequency Applications

Performance Analysis of Edge Computing in IoT

Robust fault tolerant control allocation for a modern over‐actuated commercial aircraft

Contact Info

Product

Resources

About