Table 6 Offloading Performance Comparison: we compare MLR-LC-DRLO with fine-tuned DQN, Double-DQN and CEM to show that MLR-LC-DRLO achieves better latency-critical offloading performance

Workflow Topology

   Topology 1

98.33%±0.56%

96.13%±3.24%

537±32

95.33%±1.78%

90.22±2.23%

\({\textbf {653}}\pm {\textbf {28}}\)

96.12%±0.65%

93.23±2.15%

\(611\pm 34\)

\(87.55\%\pm 2.76\%\)

\(85.31\%\pm 5.82\%\)

\(826\pm 57\)

   Topology 2

97.01%±1.28%

95.66%±2.46%

581±17

93.92%±1.78%

88.67±3.34%

\({\textbf {692}}\pm {\textbf {36}}\)

95.55%±1.38%

92.14±3.24%

\(649\pm 34\)

\(83.84\%\pm 3.57\%\)

\(80.06\%\pm 1.36\%\)

\(864\pm 39\)

   n=20

97.33%±1.14%

95.88%±2.16%

553±26

94.52%±2.63%

89.53±3.25%

\({\textbf {633}}\pm {\textbf {37}}\)

95.36%±0.91%

91.53±2.26%

\(603\pm 11\)

\(89.98\%\pm 1.36\%\)

\(87.43\%\pm 3.35\%\)

\(638\pm 42\)

   n=30

95.33%±0.34%

94.27%±1.24%

569±31

91.57%±3.21%

88.56±1.53%

\({\textbf {687}}\pm {\textbf {45}}\)

93.66%±1.80%

91.08±1.30%

\(579\pm 43\)

\(86.50\%\pm 1.09\%\)

\(85.66\%\pm 2.57\%\)

\(712\pm 35\)

   UT=DT=8.5Mbps

98.12%±0.68%

96.77%±2.15%

493±46

93.58%±2.20%

91.65±1.10%

\(586\pm 35\)

97.39%±0.60%

95.42±3.63%

\(523\pm 52\)

\(89.33\%\pm 1.55\%\)

\(86.45\%\pm 2.33\%\)

\(721\pm 22\)

   UT=DT=5.5Mbps

96.06%±1.06%

95.69%±1.53%

556±32

91.06%±2.73%

90.32±0.34%

\({\textbf {635}}\pm {\textbf {30}}\)

95.53%±1.66%

93.65±1.98%

\(609\pm 25\)

\(85.34\%\pm 3.87\%\)

\(82.68\%\pm 2.08\%\)

\(865\pm 48\)