Task Offloading Strategy With Soft Time Window In Mobile Edge Computing

In the Mobile Edge Computing(MEC)system,task offloading is a key application which has been widely concerned by scholars since its emergence.Most of scholars focused on offloading strategy under strict deadline constrain,which means use hard time window to limit the task completion time.Meanwhile,most of existing work concerned the offloading strategy from users' view instead of service providers'.In fact,the tolerance of overtime of users is elastic.They pay of task completion will decrease with time goes by due to users' tolerance change.In additional,ISPs(Internet Service Providers)also need to consider the influence of offloading strategy,realize the demand of energy saving by adjusting the processing power to gain more profit.In this paper,we designed an offloading model with soft time window,and used profit to energy consume ratio to measure the offloading efficiency.To design the full offloading strategy,we proposed a full offloading model with the maximization of profit to energy consume ratio.We applied the discrete dynamic voltage scaling and peer to peer offloading strategy between the base stations to improve offloading efficiency.To solve the problem of low local search accuracy in the combined biogeographic optimization algorithm,we introduced the crossover operator of genetic algorithm to bio-geography optimization algorithm improved local search efficiency which maintain the better solution fragment in population.Finally,we compare our algorithm to other two algorithms to ensure the performance of our algorithm.As for partial offloading strategy,to improve task processing speed and make use of local computing resource,we applied the shard method to achieve parallel processing,and we also considered the dependency between subtasks.Due to the large searching scope of problem,we introduced the multi-population cooperative searching strategy to improve searching efficiency and get better solutions.The simulation results shown our algorithm could improve the gain of per unit of energy consumption.