feat: partition by mem & node latency

[thenatlog]

Pull Request: Add Partitioning Strategy Flag with Latency-Aware Default

Description

Feature Added

• Partitioning Strategy Flag:
  • Introduced --partitioning-strategy CLI flag in main.py to allow selection between partitioning strategies.
  • Set latency_aware as the default partitioning strategy.

Implementation Details

• Imports:
  • Imported LatencyAwarePartitioningStrategy alongside existing strategies.
• Argument Parsing:
  • Updated main.py to parse the new --partitioning-strategy flag.
• Strategy Selection:
  • Instantiated the selected partitioning strategy based on user input.
• Node Initialization:
  • Modified StandardNode initialization to accept the chosen partitioning strategy.

Testing

• New Tests:
  • Added TestLatencyAwarePartitioningStrategy in tests/test_latency_aware_partitioning.py to validate the new strategy.

Additional Changes

• Protobuf Definitions:
  • Updated protobuf definitions to include latency in topology.
• gRPC Files:
  • Regenerated node_service_pb2.py and node_service_pb2_grpc.py to reflect .proto updates.

Usage

Default (Latency-Aware)

python main.py run <model_name>