Advances, Systems and Applications
Goals | Approaches that Apply Hard Constraints (elementary methods in italics) |
---|---|
1 (power only) | Deploy to the most power-efficient processor first, across all clouds (greedy on power). |
2 (capacity only) | 2-D bin-packing (for memory and processing constraints) (e.g., [1]) |
3 (delay only) | Deploy on one cloud to eliminate network delays, chosen based on delay to the user and speed of processing. |
1,2 | 2-D bin-packing (with dimensions of capacity and memory) augmented by considering power (e.g. [2]) (used in LPD). |
1,3 | Graph partitioning among clouds to control delay; for each cloud deploy to the most efficient processor first (as in LPD without capacity constraints). |
2,3 | Minimize response time (e.g. by graph partitioning accounting for delay and capacity as in [3]), and accept the solution if it meets the response time constraint. |
1,2,3 | New in LPD: Use graph partitioning methods among clouds to control delay, with two-dimensional bin-packing (capacity and memory) within clouds augmented by decisions based on power. |