Algorithm 1 PA-DRL training algorithm

Input: SFC request with the requirement of VNFs and virtual links

Output: Deployment and backup locations for each VNF of SFC

Initialize the neural network $N$ with the parameter $\omega$, and clear the replay memory buffer $D$.

while$i{s}_{end}\ne 0$do

Calculate SFC priority level $p_{sfc}$ as (6)

Calculate node priority level $p_{node}$ as (7)

Determine action set by Table 1

Select action $A\left(t\right)$ by $\epsilon -greedy$ method

Calculate reward $R\left(t\right)$ as (10)

Store the sample $\left(S\left(t\right),A\left(t\right),S\left(t+1\right),R\left(t\right),i{s}_{end}\right)$ in memory $D$

Randomly select samples from $D$ then calculate $Q$ as (14)

Train the network $N$ and update $\omega$ by using back propagation of the loss function

end while