diff --git a/Flower_v1/client_1.py b/Flower_v1/client_1.py
index ae28e95eca5c53faad1ec9e4167f959195f754b8..af26d00e8284e079351fa9bce9522c7b7248de22 100644
--- a/Flower_v1/client_1.py
+++ b/Flower_v1/client_1.py
@@ -1,4 +1,4 @@
-#python3 client_1.py <dataset> <id> <total_client> <IP:PORT>
+#python3 client_1.py <dataset> <partition> <client_num> <IP:PORT>
#python3 client_1.py cifar10 1 3 172.16.66.55:8080
import flwr as fl
@@ -10,18 +10,33 @@ import sys
# Function to load and partition the dataset based on client_id
def load_partitioned_data(dataset_name, num_clients, client_id):
if dataset_name == "cifar10":
+ '''
+ total samples train: 50000
+ total samples test: 10000
+ sample type: animals, vehicles, etc.
+ '''
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()
x_train = x_train.astype("float32") / 255.0
x_test = x_test.astype("float32") / 255.0
num_classes = 10
input_shape = (32, 32, 3)
elif dataset_name == "mnist":
+ '''
+ total samples train: 60000
+ total samples test: 10000
+ sample type: handwritten digits
+ '''
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 28, 28, 1).astype("float32") / 255.0
x_test = x_test.reshape(-1, 28, 28, 1).astype("float32") / 255.0
num_classes = 10
input_shape = (28, 28, 1)
elif dataset_name == "cifar100":
+ '''
+ total samples train: 50000
+ total samples test: 10000
+ sample type: animals, objects
+ '''
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar100.load_data()
x_train = x_train.astype("float32") / 255.0
x_test = x_test.astype("float32") / 255.0
@@ -30,7 +45,7 @@ def load_partitioned_data(dataset_name, num_clients, client_id):
else:
raise ValueError("Dataset not supported. Use 'cifar10', 'mnist', or 'cifar100'.")
- # Partition the dataset among the clients
+ # Partition the dataset among the clients (same data distribution with same number_clients setup)
total_samples = x_train.shape[0]
samples_per_client = total_samples // num_clients
@@ -39,6 +54,9 @@ def load_partitioned_data(dataset_name, num_clients, client_id):
x_client_train = x_train[start:end]
y_client_train = y_train[start:end]
+
+ # x_client_train = x_train
+ # y_client_train = y_train
# Split the client-specific data into training and validation sets
x_train, x_val, y_train, y_val = train_test_split(x_client_train, y_client_train, test_size=0.2, random_state=42)
@@ -47,7 +65,7 @@ def load_partitioned_data(dataset_name, num_clients, client_id):
# Get command-line arguments: dataset, client_id, num_clients, and server address
if len(sys.argv) != 5:
- print("Usage: python3 client.py <dataset> <partition> <client_num> <IP:PORT>")
+ print("Usage: python3 client_1.py <dataset> <partition> <client_num> <IP:PORT>")
sys.exit(1)
dataset_name = sys.argv[1]
@@ -69,17 +87,17 @@ class FederatedClient(fl.client.NumPyClient):
def fit(self, parameters, config):
model.set_weights(parameters)
- model.fit(x_train, y_train, epochs=1, batch_size=32, validation_data=(x_val, y_val))
+ model.fit(x_train, y_train, epochs=5, batch_size=32, validation_data=(x_val, y_val))
return model.get_weights(), len(x_train), {}
def evaluate(self, parameters, config):
model.set_weights(parameters)
- loss, accuracy = model.evaluate(x_test, y_test)
+ loss, accuracy = model.evaluate(x_test, y_test, verbose=0)
return loss, len(x_test), {"accuracy": float(accuracy)}
# Start Flower client
if __name__ == "__main__":
fl.client.start_client(server_address=server_address,
client=FederatedClient(),
- max_retries=5,
- max_wait_time=10)
+ max_retries=3, # retry connection to server 3 times
+ max_wait_time=5) # wait 5 seconds before retrying
diff --git a/Flower_v1/server_1.py b/Flower_v1/server_1.py
index a4664963745960faf7a038b2693c3f5b41fae720..a2ca18eb6830448f5d88bf321e87de6527e89585 100644
--- a/Flower_v1/server_1.py
+++ b/Flower_v1/server_1.py
@@ -8,7 +8,7 @@ from flwr.server.strategy import Strategy
from flwr.server.client_manager import ClientManager
from flwr.server.client_proxy import ClientProxy
from flwr.server import ServerApp, ServerConfig
-from flwr.server.strategy import FedAvg, FedMedian, FedOpt, FedAdam
+from flwr.server.strategy import FedAvg, FedMedian, FedOpt, FedAdam, FedProx, QFedAvg
import argparse
import tensorflow as tf
import flwr
@@ -110,8 +110,8 @@ parser.add_argument(
"--strategy",
type=str,
default="fedAvg",
- choices=["fedAvg", "fedMedian", "fedOpt", "fedAdam", "fedAvg2Clients"], # Add options strategies
- help="Select strategy: fedAvg, fedMedian, fedOpt or fedAdam or fedAvg2Clients",
+ choices=["fedAvg", "fedMedian", "fedOpt", "fedAdam", "fedAvg2Clients", "fedProx", "QfedAvg"], # Add options strategies
+ help="Select strategy: fedAvg, fedMedian, fedOpt, fedAdam, fedProx, QfedAvg or fedAvg2Clients",
)
parser.add_argument("-l", "--log_dir", type=str, default=None, help="Directory to save logs")
@@ -141,24 +141,46 @@ weights = model.get_weights()
parameters = flwr.common.ndarrays_to_parameters(weights)
# Select the base strategy
-if args.strategy == "fedAvg":
- base_strategy = FedAvg(evaluate_metrics_aggregation_fn=weighted_average)
-elif args.strategy == "fedAvg2Clients":
+if args.strategy == "fedAvg":
+ # baseline strategy
base_strategy = FedAvg(
- fraction_fit=0.67,
- min_fit_clients=2,
- min_available_clients=2,
+ evaluate_metrics_aggregation_fn=weighted_average)
+elif args.strategy == "fedAvg2Clients":
+ # Custom client num each round
+ base_strategy = FedAvg(
+ fraction_fit=0.67, # 67% of clients agg/fit each round
+ min_fit_clients=2, # At least 2 clients fit each round
+ min_available_clients=2, # At least 2 clients available each round
evaluate_metrics_aggregation_fn=weighted_average,
)
elif args.strategy == "fedMedian":
- base_strategy = FedMedian(evaluate_metrics_aggregation_fn=weighted_average)
-elif args.strategy == "fedOpt":
+ # robustness strategy
+ base_strategy = FedMedian(
+ evaluate_metrics_aggregation_fn=weighted_average)
+elif args.strategy == "fedOpt":
+ # server optimization strategy
base_strategy = FedOpt(
- evaluate_metrics_aggregation_fn=weighted_average, initial_parameters=parameters
+ evaluate_metrics_aggregation_fn=weighted_average,
+ initial_parameters=parameters
)
elif args.strategy == "fedAdam":
+ # server optimization strategy use Adam
base_strategy = FedAdam(
- evaluate_metrics_aggregation_fn=weighted_average, initial_parameters=parameters
+ evaluate_metrics_aggregation_fn=weighted_average,
+ initial_parameters=parameters
+ )
+elif args.strategy == "fedProx":
+ # regularization strategy
+ base_strategy = FedProx(
+ evaluate_metrics_aggregation_fn=weighted_average,
+ initial_parameters=parameters,
+ proximal_mu = 0.1
+ )
+elif args.strategy == "QFedAvg":
+ # improve fairness, reduce var_loss strategy
+ base_strategy = QFedAvg(
+ evaluate_metrics_aggregation_fn=weighted_average,
+ initial_parameters=parameters
)
diff --git a/README.md b/README.md
index 40440191c8f60266f093ce561925a950d721ea4f..62c3af3012881a58bf46ebd191abdc2cfdd165c2 100644
--- a/README.md
+++ b/README.md
@@ -9,7 +9,7 @@ This project provides tools to measure the energy consumption of Flower-based fe
- [Usage](#usage)
- [FL framework](#fl-framework)
- [Configure instances for CPU](#configure-instance-for-cpu)
- - [Configure instances for GPU](#configure-instance-for-gpu)
+ - [Configure instances for GPU](#configure-instance-for-gpu-developing)
- [Run exp](#run-exp)
- [Quickstart](#quickstart)
- [Step 1. Reserve the Hosts in G5K](#step-1-reserve-the-hosts-in-g5k)
@@ -69,7 +69,7 @@ Configure instances of experiment in a json format, structure is shown below.
```bash
bash "$(python3 -c "import expetator, os; print(os.path.join(os.path.dirname(expetator.__file__), 'leverages', 'dvfs_pct.sh'))")" init
```
- - Run this command to get available frequencies:
+ - Run this command to get available 4 frequencies (min, max, 2 in middle):
```bash
python3 get_frequencies.py
```
@@ -114,15 +114,26 @@ Configure instances of experiment in a json format, structure is shown below.
}
```
-### Configure instance for GPU
+### Configure instance for GPU (*developing)
- The configuration is as same CPU, except **dvfs** role. In GPU config, the role is **dvfs_gpu**.
-Choose only one in 3 settings (steps - zoomfrom - zoomto use for one setting).
+Choose only one in 3 settings:
- `dummy`: for testing in min and max GPU freq (`false` or `true`).
- `baseline`: for testing in max GPU freq (`false` or `true`).
- - `steps`: steps to jump in range/window of frequencies (int).
- - `zoomfrom`: freq start
- - `zoomto`: freq stop
+ - last setting includes 3 parameters: test with the all freqs in the range. To disable this setting, set the `zoomfrom` and `zoomto` are same values.
+ - `steps`: steps to jump in range/window of frequencies (int).
+ - `zoomfrom`: freq start
+ - `zoomto`: freq stop
+ *example:*
+ - list of freq available [1, 1.1, 1.2, 1.9, 2.5, 2.7] GHz
+ - with `zoomfrom` = 1.1, `zoomto` = 2.7 and `steps` = 2
+ - list of tested freq returns [1.1, 1.9, 2,7]
+
+ **Remark:** check the available frequencies before using option `frequencies`.
+ Run below cmd:
+ ```bash
+ nvidia-smi -i 0 --query-supported-clocks=gr --format=csv,noheader,nounits | tr '\n' ' '
+ ```
```json
"dvfs_gpu": {
@@ -169,6 +180,8 @@ Reserve 4 hosts (GPU) (1 server + 3 clients) for 2 hours:
oarsub -I -t exotic -p "gpu_count>0" -l {"cluster='drac'"}/host=4 # grenoble
```
+**Remark**: for now only 2 clusters, `chifflot` in Lille and `drac` in Grenoble are available for testing in more than 3 GPU nodes, maximum is 8 (`chifflot`) or 12 (`drac`) nodes.
+
Make sure your are in`eflwr/Run/`:
```bash
cd Run
diff --git a/Run/config_instances.json b/Run/config_instances.json
index 941ec5d3072819fbac2bca1023278192cd5391e3..b8bac0b373d5ca64667d26c2d6bbab31343db06a 100644
--- a/Run/config_instances.json
+++ b/Run/config_instances.json
@@ -4,7 +4,7 @@
"instance": "fedAvg_cifar10",
"output_dir": "./Log",
"dvfs_cpu": {
- "dummy": true,
+ "dummy": false,
"baseline": false,
"frequencies": null
},
@@ -26,7 +26,7 @@
"args": [
"./Flower_v1/client_1.py",
"cifar10",
- "1",
+ "0",
"3"
],
"ip": "172.16.66.77"
@@ -37,7 +37,7 @@
"args": [
"./Flower_v1/client_1.py",
"cifar10",
- "2",
+ "1",
"3"
],
"ip": "172.16.66.78"
@@ -48,7 +48,7 @@
"args": [
"./Flower_v1/client_1.py",
"cifar10",
- "3",
+ "2",
"3"
],
"ip": "172.16.66.79"
@@ -56,6 +56,116 @@
]
},
"2": {
+ "instance": "fedAvg_cifar100",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedAvg"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "3": {
+ "instance": "fedAvg_mnist",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedAvg"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "4": {
"instance": "fedAvg2Clients_cifar10",
"output_dir": "./Log",
"dvfs_cpu": {
@@ -78,12 +188,177 @@
{
"name": "client1",
"command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar10",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
"args": [
"./Flower_v1/client_1.py",
"cifar10",
"1",
"3"
],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar10",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "5": {
+ "instance": "fedAvg2Clients_cifar100",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedAvg2Clients"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "6": {
+ "instance": "fedAvg2Clients_mnist",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedAvg2Clients"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "7": {
+ "instance": "fedProx_cifar10",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedProx"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar10",
+ "0",
+ "3"
+ ],
"ip": "172.16.66.77"
},
{
@@ -92,18 +367,293 @@
"args": [
"./Flower_v1/client_1.py",
"cifar10",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar10",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "8": {
+ "instance": "fedProx_cifar100",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedProx"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
"2",
"3"
],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "9": {
+ "instance": "fedProx_mnist",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedProx"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "1",
+ "3"
+ ],
"ip": "172.16.66.78"
},
{
"name": "client3",
"command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "10": {
+ "instance": "fedAdam_cifar10",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedAdam"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar10",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
"args": [
"./Flower_v1/client_1.py",
"cifar10",
- "3",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar10",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "11": {
+ "instance": "fedAdam_cifar100",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedAdam"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "cifar100",
+ "2",
+ "3"
+ ],
+ "ip": "172.16.66.79"
+ }
+ ]
+ },
+ "12": {
+ "instance": "fedAdam_mnist",
+ "output_dir": "./Log",
+ "dvfs_cpu": {
+ "dummy": true,
+ "baseline": false,
+ "frequencies": null
+ },
+ "server": {
+ "command": "python3",
+ "args": [
+ "./Flower_v1/server_1.py",
+ "-r 1",
+ "-s fedAdam"
+ ],
+ "ip": "172.16.66.76",
+ "modules": ["logger"],
+ "port": 8080
+ },
+ "clients": [
+ {
+ "name": "client1",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "0",
+ "3"
+ ],
+ "ip": "172.16.66.77"
+ },
+ {
+ "name": "client2",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "1",
+ "3"
+ ],
+ "ip": "172.16.66.78"
+ },
+ {
+ "name": "client3",
+ "command": "python3",
+ "args": [
+ "./Flower_v1/client_1.py",
+ "mnist",
+ "2",
"3"
],
"ip": "172.16.66.79"
diff --git a/Run/custom_gpuclock.py b/Run/custom_gpuclock.py
index 952efb259a0c52d402e5cd45ad2ff0c8a2b1982b..4046de7449e9b6ef3de4bae00d6561b77cbf71e4 100644
--- a/Run/custom_gpuclock.py
+++ b/Run/custom_gpuclock.py
@@ -1,10 +1,7 @@
-from expetator.leverages.gpuclock import GpuClock
-class CustomGpuClock(GpuClock):
- """ Custom GPU Clock Leverage using clocks.gr instead of clocks.applications.gr """
-
+class GpuClock:
+ 'nvidia-smi graphics/sm clock setting'
def __init__(self, dummy=False, baseline=False, steps=2, zoomfrom=0, zoomto=0):
- super().__init__()
self.dummy = dummy
self.baseline = baseline
self.executor = None
@@ -12,74 +9,62 @@ class CustomGpuClock(GpuClock):
self.available_frequencies_mem = []
self.clock_mem_max = None
self.clock_sm_min = None
- self.nsteps = steps
-
- # Zoom enables fine granularity within a frequency window
+ self.nsteps = steps
+ # zoomfrom/to enables full freq. granularity within a clock window:
if zoomto != 0 and zoomfrom != zoomto:
self.zoom = (zoomfrom, zoomto)
else:
self.zoom = None
def build(self, executor):
- """ Gather the available frequencies """
+ 'Gather the available frequencies'
self.executor = executor
q = "nvidia-smi -i 0 --query-supported-clocks=gr --format=csv,noheader,nounits | tr '\n' ' '"
clk_s = self.executor.local(q)
- clk = sorted([int(f) for f in clk_s.strip().split() if f.isdigit()])
-
- if not clk:
- raise RuntimeError("Failed to retrieve supported GPU clock frequencies.")
-
+ clk = sorted([int(f) for f in clk_s.strip().split(' ')])
self.clock_sm_min = clk[0]
- self.available_frequencies = [clk[(i * (len(clk) - 1)) // (self.nsteps - 1)] for i in range(self.nsteps)]
-
+ self.available_frequencies = [clk[(i*(len(clk)-1))//(self.nsteps-1)] for i in range(self.nsteps)]
q = "nvidia-smi -i 0 --query-supported-clocks=mem --format=csv,noheader,nounits | tr '\n' ' '"
clk_s = self.executor.local(q)
- self.available_frequencies_mem = sorted([int(f) for f in clk_s.strip().split() if f.isdigit()])
+ self.available_frequencies_mem = sorted([int(f) for f in clk_s.strip().split(' ')])
self.clock_mem_max = self.available_frequencies_mem[-1]
-
- if self.zoom:
- clkz = [f for f in clk if self.zoom[0] <= f <= self.zoom[1]]
- rest = [f for f in self.available_frequencies if f < self.zoom[0] or f > self.zoom[1]]
+ if self.zoom is not None:
+ # pick all available freq within the zoom window:
+ clkz = [f for f in clk if self.zoom[0] <= f and f <= self.zoom[1]]
+ rest = [f for f in self.available_frequencies if f < self.zoom[0] or self.zoom[1] < f]
self.available_frequencies = sorted(clkz + rest)
-
if self.dummy:
self.available_frequencies = [self.available_frequencies[0], self.available_frequencies[-1]]
-
if self.baseline:
self.available_frequencies = [self.available_frequencies[-1]]
def available_states(self):
- """ Returns all available frequencies """
+ 'Returns all the frequencies'
return self.available_frequencies
def start(self, freq):
- """ Sets the GPU frequency """
+ 'Sets the right frequency range on gpu 0'
if freq in self.available_frequencies:
- self.executor.local(f'nvidia-smi -i 0 -ac {self.clock_mem_max},{freq}', root=True)
+ self.executor.local('nvidia-smi -i 0 -ac %s,%s' % (self.clock_mem_max, freq), root=True)
+ #self.executor.local('nvidia-smi -i 0 -lgc %s,%s' % (self.clock_sm_min, freq), root=True)
def stop(self, output_file=None):
- """ Reset GPU to default frequency """
+ 'Reverts to the maximum frequency'
self.executor.local('nvidia-smi -i 0 -rac', root=True)
-
+ #self.executor.local('nvidia-smi -i 0 -rgc', root=True)
+
def get_state(self):
- """ Returns the current min and max application frequencies using clocks.gr instead of clocks.applications.gr """
+ 'Returns the current min and max application frequencies'
cur_min = self.clock_sm_min
- cur_max_str = self.executor.local('nvidia-smi -i 0 --query-gpu=clocks.gr --format=csv,noheader,nounits').strip()
-
- try:
- cur_max = int(cur_max_str)
- except ValueError:
- cur_max = 0 # Fallback in case of an error
-
- return cur_min, cur_max, self.clock_mem_max
+ cur_max = int(self.executor.local('nvidia-smi -i 0 --query-gpu=clocks.applications.gr --format=csv,noheader,nounits'))
+ return (cur_min, cur_max, self.clock_mem_max)
def state_to_str(self):
- """ Returns the current min and max frequencies as a string """
- cur_min, cur_max, mem_max = self.get_state()
- return f'{cur_min} {cur_max} {mem_max}'
+ 'Returns the current min and max frequencies as a string'
+ (cur_min, cur_max, mem_max) = self.get_state()
+ return '%s %s %s' % (cur_min, cur_max, mem_max)
def get_labels(self):
- """ Returns labels for frequencies """
+ 'Returns the label for frequencies'
return ('fmin', 'fmax', 'fmemax')