To learn more about the energy cost of data movement on mobile platforms (where energy is scarce), I read the following recent paper about the topic:
D. Pandiyan and C.J. Wu, Quantifying the energy cost of data movement for emerging smart phone workloads on mobile platforms, in Proceedings of the IEEE International Symposium on Workload Characterization, 2014. (Available at http://ieeexplore.ieee.org/xpl/articleDetails.jsp?reload=true&arnumber=6983056 in CMU network.)
The essential conclusion is that a staggering 34.6% (up to 41%) of total energy (excluding peripherals like screen & wifi) is spent on moving data from one level of the memory hierarchy to the next level! And, the experiment by Pandiyan et al. is done on Samsung Galaxy S3 device running Samsung-made Exynos4 Quad 4412 SoC with Android 4.3. From this, it is really apparent that locality can not only improve the performance of programs but also conserve energy, which is crucial for data centers and mobile devices.
grarawr
Is memory movement the dominating factor in performance cost?
bdebebe
@grarawr The dominating factor in performance cost isn't as easy to pin down as is the energy cost of data movement being so dominant, but it is definitely one of the leading causes we learned about in this class.
To learn more about the energy cost of data movement on mobile platforms (where energy is scarce), I read the following recent paper about the topic:
D. Pandiyan and C.J. Wu, Quantifying the energy cost of data movement for emerging smart phone workloads on mobile platforms, in Proceedings of the IEEE International Symposium on Workload Characterization, 2014. (Available at http://ieeexplore.ieee.org/xpl/articleDetails.jsp?reload=true&arnumber=6983056 in CMU network.)
The essential conclusion is that a staggering 34.6% (up to 41%) of total energy (excluding peripherals like screen & wifi) is spent on moving data from one level of the memory hierarchy to the next level! And, the experiment by Pandiyan et al. is done on Samsung Galaxy S3 device running Samsung-made Exynos4 Quad 4412 SoC with Android 4.3. From this, it is really apparent that locality can not only improve the performance of programs but also conserve energy, which is crucial for data centers and mobile devices.
Is memory movement the dominating factor in performance cost?
@grarawr The dominating factor in performance cost isn't as easy to pin down as is the energy cost of data movement being so dominant, but it is definitely one of the leading causes we learned about in this class.