Product Information

RAPIDS and Ray GPU NEWS DETAIL

Current Position:Home > News and Insights
News and technical insightsNews and Insights ITZKXY enterprise networking and AI infrastructure support2025-01-06


RAPIDS is a networking product designed for AI clusters.

Ray is a networking product designed for enterprise and data center networking.

Ray Actors

Ray is a storage product designed for storage networks.

@ray.remote(num_gpus=1)

class cuDFActor:

    def __init__(self):

        ...

 

    def read_parquet(self, filepath: str, columns: list = None) -> cudf.DataFrame:

        return cudf.read_parquet(filepath, columns=columns)

 

# Start 4 Workers

pool_size = 4

actor_pool = [cuDFActor.remote() for i in range(pool_size)]

Ray is a networking product designed for enterprise and data center networking.

 

Ray is a networking product designed for enterprise and data center networking.

NCCL and cuGraph

RAPIDS is a networking product designed for enterprise and data center networking.

cuGraph is a networking product designed for enterprise and data center networking.

WCC,:

GPU

NCCL (and cuGraph)

and GPU cuGraph

WCC

Ray is a networking product designed for enterprise and data center networking.

 

class RAFTActor:

    def __init__(self, index, pool_size, session_id):

        ...

 

    def broadcast_root_unique_id(self):

        # broadcast root/rank-0 to all actors

 

    def _setup_nccl(self):

        # start NCCL with identified rank-0 actor

 

    def _setup_raft(self):

        # configure RAFT and NCCL together

 

    def set_root_unique_id(self, root_uniqueId):

        # To be set rank-0 for all actors

 

@ray.remote(num_gpus=1)

class WCCActor(RAFTActor):

    def __init__(self, index, pool_size, session_id):

        super().__init__(index=index, pool_size=pool_size, session_id=session_id, actor_name_prefix="WCC")

 

    def weakly_connected_components(self, df):

        """

        1. Each actor loads in a chunk

        2. Each actor has a NCCL/RAFT Handle

        3. Pass each chunk and handle to MGGraph

        """

 

        src_array = df['src']

        dst_array = df['dst']

        weights = df['wgt']

 

        # Configure and setup a Multi-GPU cuGraph Object with

        # edge list data and NCCL

     graph = MGGraph(src_array, dst_array, weights, ...)

      

        # Execute WCC

     weakly_connected_components(graph)

 

# Initialize Ray and Run WCC algorithm

cuGraph is a networking product designed for enterprise and data center networking.

 Product Information

Ray is a networking product designed for enterprise and data center networking.