Introduction
Related work
The time-sensitive learning-to-rank approach for resource prediction
Parameters | Indicators | Description |
---|---|---|
Pre-execution parameters | CPU core | Number of CPU cores used |
Entity | Number of simulation entities | |
Events | Number of simulation events executed | |
Lookahead | Restricts the simulation events that can be executed between the current timestamp and the sum of timestamp and lookahead values | |
Simulation time | The virtual time in simulation applications | |
Runtime parameters | CPU usage | Average CPU usage |
CPU usage_max | CPU usage maximum | |
CPU usage_min | CPU usage minimum | |
Memory | Average memory consumption | |
Memory_max | Memory consumption maximum | |
Memory_min | Memory consumption maximum | |
File | Hard drive read/write capacity | |
Network rec | Network upload rate | |
Network rec_max | Maximum network upload rate | |
Network rec_min | Minimum network upload rate | |
Network sent | Network download rate | |
Network sent_max | Maximum network download rate | |
Network sent_min | Minimum network download rate | |
Network delay | Network latency |
The LTR model
The time-sensitive LTR method for resource prediction
Evaluation metrics
Experimental results and analysis
Application and experimental settings
Simulation application | Features | Values |
---|---|---|
Phold | CPU cores | [1, 2, 4, 8, 16, 24, 32] |
Entity | [1000, 2000, 3000, 4000, 5000] | |
Event | [100, 200, 300, 400, 500] | |
Simulation time | [2000] | |
Lookahead | [0.2, 0.4, 0.6, 0.8. 1.0] | |
PODM | CPU cores | [1, 2, 4, 8, 16, 24, 32] |
Individuals | [500, 1000, 1500, 2000] | |
Hot events | [5,10,20] | |
Cities | [10,20,30] | |
Media | [10] | |
Simulation time | [1000] | |
Lookahead | [0.1, 0.5, 1.0] |
Feature extraction method based on the SHAP interpretable framework
Model parameter setting and sensitivity analysis
Model | Parameters |
---|---|
ETR | Num of trees = 200 |
LR | Max iterations = 100, regularization parameter = 0.2 |
RF | Max depth = 4, num of trees = 300 |
XGB | Learning rate = 0.05, max depth = 3, Max iterations = 300 |
MLP | Hidden layer = (200,80), learning rate = 0.1, max iterations = 600 |
SVR | Kernel = “linear”, regularization parameter = 0.1, max iterations = 500 |
Rankboost | Learning rate = 0.15, max depth = 5, max iterations = 800 |
RBM | Learning rate = 0.2, max iterations = 400 |
Model performance evaluation
Run-time comparison
Model | Training time | Ranking result output time |
---|---|---|
RBM | 60.325 | 3.15 |
SVR | 1.165 | 1.44 |
MLP | 55.101 | 2.52 |
LR | 1.035 | 1.74 |
ETR | 2.012 | 1.23 |
RF | 2.023 | 2.58 |
XBG | 30.267 | 3.25 |
Rankboost | 86.683 | 0.52 |
TSLTR | 116.04 | 0.55 |
Simulation application | CPU core usage (Listed in descending order of prediction results) | Actual runtime (s) |
---|---|---|
Phold_1 (simulation entities are set 3000) | 24 | 456 |
32 | 910 | |
16 | 1075 | |
8 | 1845 | |
4 | 2628 | |
2 | 3428 | |
1 | 6294 | |
Phold_2 (simulation entities are set 1000) | 16 | 928 |
24 | 1966 | |
32 | 1083 | |
8 | 3045 | |
4 | 5742 | |
2 | 10,578 | |
1 | 20,253 |