[GRAPHENE] Implement Graphene Scheduler (#91)

* [GRAPHENE] Begin implementation of the Graphene scheduler.

* Implement basic structure of Graphene's two-phase scheduling.

* Implement a separate STRL session to invoke min makespan scheduling.

* Add a method to update the TaskGraph in-place.

* Test for imports of both Graphene and TetriSched schedulers together.

* Finish the implementation of the offline pass of Graphene.

* Implement the example Workload from Graphene paper.

* Fix black formatting for workload/tasks.py

* Fix flake8 for Graphene.

* Introduce notify_workload_updated to transform TaskGraphs before Tasks from them are released.

configs/simple_graphene_workload.conf

@@ -0,0 +1,16 @@
+# Output configs
+--log=./simple_graphene_workload.log
+--log_level=debug
+--csv=./simple_graphene_workload.csv
+
+# Task configs
+--runtime_variance=0
+
+# Scheduler configs
+--scheduler=Graphene
+--scheduler_runtime=0
+
+# Execution mode configs
+--execution_mode=json
+--workload_profile_path=./profiles/workload/simple_graphene_workload.yaml
+--worker_profile_path=./profiles/workers/simple_graphene_workers.yaml

main.py

@@ -19,6 +19,7 @@
     ClockworkScheduler,
     EDFScheduler,
     FIFOScheduler,
+    GrapheneScheduler,
     ILPScheduler,
     LSFScheduler,
     TetriSchedCPLEXScheduler,
@@ -289,6 +290,7 @@
         "Clockwork",
         "TetriSched",
         "GraphenePrime",
+        "Graphene",
     ],
     "The scheduler to use for this execution.",
 )
@@ -841,6 +843,20 @@ def main(args):
             log_to_file=FLAGS.scheduler_log_to_file,
             _flags=FLAGS,
         )
+    elif FLAGS.scheduler == "Graphene":
+        scheduler = GrapheneScheduler(
+            preemptive=FLAGS.preemption,
+            runtime=EventTime(FLAGS.scheduler_runtime, EventTime.Unit.US),
+            lookahead=EventTime(FLAGS.scheduler_lookahead, EventTime.Unit.US),
+            retract_schedules=FLAGS.retract_schedules,
+            goal=FLAGS.ilp_goal,
+            time_discretization=EventTime(
+                FLAGS.scheduler_time_discretization, EventTime.Unit.US
+            ),
+            plan_ahead=EventTime(FLAGS.scheduler_plan_ahead, EventTime.Unit.US),
+            log_to_file=FLAGS.scheduler_log_to_file,
+            _flags=FLAGS,
+        )
     else:
         raise ValueError(
             "Unsupported scheduler implementation: {}".format(FLAGS.scheduler)

profiles/workers/simple_graphene_workers.yaml

@@ -0,0 +1,6 @@
+- name: WorkerPool_1
+  workers:
+      - name: Worker_1_1
+        resources:
+            - name: Slot
+              quantity: 3

profiles/workload/simple_graphene_workload.yaml

@@ -0,0 +1,73 @@
+# Implementation of the example workload specified in the Graphene paper.
+# This workload is being used to test the Offline and Online stages of the
+# ILP-based reimplementation of the Graphene algorithm.
+graphs:
+  - name: GrapheneMotivation
+    graph:
+      - name: "Task0"
+        work_profile: "Task0Profile"
+        children: ["Task6"]
+      - name: "Task1"
+        work_profile: "Task1Profile"
+        children: ["Task2"]
+      - name: "Task2"
+        work_profile: "Task2Profile"
+        children: ["Task6"]
+      - name: "Task3"
+        work_profile: "Task3Profile"
+        children: ["Task4"]
+      - name: "Task4"
+        work_profile: "Task4Profile"
+        children: ["Task5"]
+      - name: "Task5"
+        work_profile: "Task5Profile"
+        children: ["Task6"]
+      - name: "Task6"
+        work_profile: "Task6Profile"
+    release_policy: fixed
+    period: 0 # In us for now.
+    invocations: 1
+    deadline_variance: [2000, 2000]
+profiles:
+  - name: Task0Profile
+    execution_strategies:
+      - batch_size: 1
+        runtime: 100
+        resource_requirements:
+          Slot:any: 1
+  - name: Task1Profile
+    execution_strategies:
+      - batch_size: 1
+        runtime: 20
+        resource_requirements:
+          Slot:any: 3
+  - name: Task2Profile
+    execution_strategies:
+      - batch_size: 1
+        runtime: 100
+        resource_requirements:
+          Slot:any: 1
+  - name: Task3Profile
+    execution_strategies:
+      - batch_size: 1
+        runtime: 20
+        resource_requirements:
+          Slot:any: 3
+  - name: Task4Profile
+    execution_strategies:
+      - batch_size: 1
+        runtime: 20
+        resource_requirements:
+          Slot:any: 3
+  - name: Task5Profile
+    execution_strategies:
+      - batch_size: 1
+        runtime: 100
+        resource_requirements:
+          Slot:any: 1
+  - name: Task6Profile
+    execution_strategies:
+      - batch_size: 1
+        runtime: 20
+        resource_requirements:
+          Slot:any: 1

schedulers/__init__.py

@@ -12,7 +12,9 @@
 from .tetrisched_gurobi_scheduler import TetriSchedGurobiScheduler
 
 try:
+    from .graphene_scheduler import GrapheneScheduler
     from .tetrisched_scheduler import TetriSchedScheduler
 except ImportError:
     pass
+
 from .z3_scheduler import Z3Scheduler

schedulers/base_scheduler.py

@@ -179,6 +179,25 @@ def start(
             placements=placements,
         )
 
+    def notify_workload_updated(
+        self,
+        sim_time: EventTime,
+        workload: Workload,
+        worker_pools: "WorkerPools",  # noqa: F821
+    ) -> None:
+        """Notifies the Scheduler that the Workload has been updated.
+
+        The Simulator invokes this method to notify the Scheduler that it can do
+        any necessary bookkeeping or updates to its internal state.
+
+        Args:
+            sim_time (`EventTime`): The time at which the workload was updated.
+            workload (`Workload`): The `Workload` that was updated.
+            worker_pools (`WorkerPools`): The set of worker pools available to the
+                Scheduler at the update of the Workload.
+        """
+        pass
+
     def schedule(
         self,
         sim_time: EventTime,