Update README

Author: mental2008

README.md

@@ -1,3 +1,8 @@
+# 🚀 Kubernetes Scheduler Simulator
+
+The simulator evaluates different scheduling policies in GPU-sharing clusters.
+It includes the Fragmentation Gradient Descent (FGD) policy proposed in the ATC paper (Beware of Fragmentation: Scheduling GPU-Sharing Workloads with Fragmentation Gradient Descent), along with other baseline policies. 
+
 ## 🚧 Environment Setup
 
 Please ensure that Go is installed.
@@ -16,31 +21,32 @@ $ make
 
 ## 🔥 Quickstart Example
 
-The following example will schedule 6 test pods onto a cluster with 2 nodes, and the expected output will show the allocation ratio of each resource dimension (CPU, memory, GPU).
+The following example will schedule 6 pods to a cluster with 2 nodes, and the expected output will show the allocation ratio of each resource dimension (CPU, memory, GPU).
+The default scheduling policy is fragmentation gradient descent (FGD).
 
 ```bash
 $ bin/simon apply --extended-resources "gpu" \
                   -f example/test-cluster-config.yaml \
                   -s example/test-scheduler-config.yaml
 ```
 
-## 🔮 Experiments
+## 🔮 Experiments on Production Traces
 
-TBD.
+Please refer to [README](data/README.md) under the `data` directory to prepare production traces, and then refer to [README](experiments/README.md) under the `experiments` directory to reproduce the results reported in the paper.
 
 ## 📝 Paper
 
 Please cite our paper if it is helpful to your research.
 
 ```
 @inproceedings{FGD,
-title = {Beware of Fragmentation: Scheduling GPU-Sharing Workloads with Fragmentation Gradient Descent},
-author = {Qizhen Weng and Lingyun Yang and Yinghao Yu and Wei Wang and Xiaochuan Tang and Guodong Yang and Liping Zhang},
-booktitle = {2023 {USENIX} Annual Technical Conference},
-year = {2023},
-series = {{USENIX} {ATC} '23}
-url = {https://www.usenix.org/conference/atc23/presentation/weng},
-publisher = {{USENIX} Association},
+    title = {Beware of Fragmentation: Scheduling GPU-Sharing Workloads with Fragmentation Gradient Descent},
+    author = {Qizhen Weng and Lingyun Yang and Yinghao Yu and Wei Wang and Xiaochuan Tang and Guodong Yang and Liping Zhang},
+    booktitle = {2023 {USENIX} Annual Technical Conference},
+    year = {2023},
+    series = {{USENIX} {ATC} '23}
+    url = {https://www.usenix.org/conference/atc23/presentation/weng},
+    publisher = {{USENIX} Association},
 }
 ```
 

experiments/README.md

@@ -1,6 +1,7 @@
-## Steps
+## Experiment Steps
 
 ### Generate Run Scripts
+
 ```bash
 cd kubernetes-scheduler-simulator/experiments
 python run_scripts/generate_run_scripts.py > run_scripts/run_scripts_0511.sh
@@ -13,6 +14,7 @@ Script parameters:
 - NUM_REPEAT: repetited experiments (10 in paper)
 
 ### Execute
+
 ```bash
 # pwd: kubernetes-scheduler-simulator/experiments
 cd ..
@@ -27,6 +29,7 @@ cat experiments/run_scripts/run_scripts_0511.sh | while read i; do printf "%q\n"
 - 16 hours for 1020 experiments on a 256 vCPU machine with PARALLEL=128.
 
 ### Merge
+
 ```bash
 # pwd: kubernetes-scheduler-simulator/experiments
 ln -s 2023_0511 data # softlink it to data
@@ -35,6 +38,7 @@ bash merge_bash.sh
 ```
 
 ### Plot
+
 ```bash
 # pwd: kubernetes-scheduler-simulator/experiments
 ln -s analysis/analysis_results/* plot/ # softlink all csv under analysis_results/ to plot/
@@ -46,5 +50,4 @@ python plot_paib_gpushare_alloc_bar.py # Fig. 11
 python plot_paib_multigpu_alloc_bar.py # Fig. 12
 python plot_paib_gpuspec_alloc_bar.py  # Fig. 13
 python plot_paib_nongpu_alloc_bar.py   # Fig. 14
-```
-
+```