Advances, Systems and Applications
From: Deep Reinforcement Learning-Based Workload Scheduling for Edge Computing
i | index of a task |
∂ i | input data size (in bits) of taski |
β i | output data size (in bits) of taski |
c i | total number of CPU cycles that isrequired to complete the taski |
Ï„ i | delay constraint of taski |
\(t_{{com}}^{i}\) | communication delay of taski |
\(t_{{td}}^{i}\) | transmission delay of taski |
\(t_{{up}}^{i}\) | upload time of taski |
\(t_{{down}}^{i}\) | download time of taski |
\(t_{{wait}}^{i}\) | waiting time between the uploading to local edge server and starting execution |
ω wlan | bandwidth of WLAN |
p n | transmission powers of mobile device |
h n,s | channel gains between mobile device and edge server |
N 0 | noise power |
f l | local edge server computing power |
f nb | nearby edge server computing power |
f c | cloud computing power |
R man | MAN transmission rate |
\(R_{{wlan}}^{}\) | WLAN transmission rate |
λ1,…λm+1 | scheduling decision variables |
server t | states of all servers at step t |
u n | VM utilization of nth server |
network t | states of networks at step t |
T t | total delay of current task |