This dataset supports the article entitled "Predictive Thermal Management for Energy-efficient Execution of Concurrent Applications on Heterogeneous Multi-cores" accepted for publication in Transactions on Very Large Scale Integration (VLSI) Systems, January, 2019.

============================================================================

Person responsible for collecting the data: Cédric de Bellefroid and Karunakar Reddy Basireddy, krb1g15@ecs.soton.ac.uk

============================================================================

Date of data collection: 20/06/2017 to 1/08/2017 and 1/7/2018 to 6/8/2018

============================================================================

Licenses/restrictions placed on the data: CC-BY/Public - No restriction

============================================================================

Links to publications that cite or use the data: xxxxx 

============================================================================

Data Supporting Figures:

Fig. 1: Linux ondemand governor Frequency setting over time.

Fig. 2: Frequency over time with a State-of-the-art RTM.

Fig. 3: Frequency setting over time by avoiding throttling.

Fig. 5: Results from leakage coefficient computation and simulation. The predicted leakage power graph is close to the measured power of each cluster.

Fig. 6: Temperature measurements for the predicted temperature versus the actual readings and the error between them.

Fig. 7: Online comparison of different predictors using the DTM with a temperature threshold of 60, 70, 80 and 90 ◦ C.

Fig. 8: Comparison of the proposed temperature predictor with the GSA precictor [16] for different applications.

Fig. 9: Average time and energy improvements of the proposed PDTPM compared to existing approaches. Figure shows results for single (a), double (b) and triple (c) application scenarios, respectively.

Fig. 10: Average power for the evaluated scenarios with single, double and triple applications.

Fig. 11: Performance and energy improvements of the proposed PDTPM over Linux HMP + the performance governor.

Fig. 12: Performance and energy improvements of the proposed PDTPM over Linux HMP + the performance governor for double application scenarios.

Fig. 13: Performance and energy improvements of the proposed PDTPM over Linux HMP + the performance governor for triple application scenarios.

Fig. 14: Temperature variation average for the proposed associated with the different predictors. Thermal cycling is represented by the difference of temperature between two measurements of one second.

============================================================================