def test_rdm_pingpong(cmdline_args, iteration_type, completion_type, memory_type): from efa.efa_common import efa_run_client_server_test efa_run_client_server_test(cmdline_args, "fi_rdm_pingpong", iteration_type, completion_type, memory_type, "all")
def test_rdm_tagged_bw_range(cmdline_args, completion_type, memory_type, message_size): from efa.efa_common import efa_run_client_server_test efa_run_client_server_test(cmdline_args, "fi_rdm_tagged_bw", "short", completion_type, memory_type, message_size)
def test_rma_bw(cmdline_args, iteration_type, operation_type, completion_type, memory_type): from efa.efa_common import efa_run_client_server_test command = "fi_rma_bw -e rdm" command = command + " -o " + operation_type efa_run_client_server_test(cmdline_args, command, iteration_type, completion_type, memory_type, "all")
def test_runt_read_functional(cmdline_args): """ Verify runt reading protocol is working as expected by sending 1 message of 256 KB. 64 KB of the message will be transfered using EFA device's send capability The remainer will be tranfered using EFA device's RDMA read capability """ import copy from efa.efa_common import efa_run_client_server_test, efa_retrieve_hw_counter_value, has_gdrcopy cmdline_args_copy = copy.copy(cmdline_args) if cmdline_args_copy.environments: cmdline_args_copy.environments += " " else: cmdline_args_copy.environments = "" cmdline_args_copy.environments += "FI_EFA_USE_DEVICE_RDMA_READ=1 FI_EFA_RUNT_SIZE=65536 FI_HMEM_CUDA_USE_GDRCOPY=1" # currently, runting read is enabled only if gdrcopy is available. # thus skip the test if gdrcopy is not available if not has_gdrcopy(cmdline_args.server_id) or not has_gdrcopy( cmdline_args.client_id): pytest.skip("No gdrcopy") return # wrs stands for work requests server_read_wrs_before_test = efa_retrieve_hw_counter_value( cmdline_args.server_id, "rdma_read_wrs") server_read_bytes_before_test = efa_retrieve_hw_counter_value( cmdline_args.server_id, "rdma_read_bytes") client_send_bytes_before_test = efa_retrieve_hw_counter_value( cmdline_args.client_id, "send_bytes") # currently runting read is only used on cuda memory, hence set the memory_type to "cuda_to_cuda" efa_run_client_server_test(cmdline_args_copy, "fi_rdm_tagged_bw", iteration_type="1", completion_type="transmit_complete", memory_type="cuda_to_cuda", message_size="262144", warmup_iteration_type="0") server_read_wrs_after_test = efa_retrieve_hw_counter_value( cmdline_args.server_id, "rdma_read_wrs") server_read_bytes_after_test = efa_retrieve_hw_counter_value( cmdline_args.server_id, "rdma_read_bytes") client_send_bytes_after_test = efa_retrieve_hw_counter_value( cmdline_args.client_id, "send_bytes") print("server_read_bytes_before_test: {}".format( server_read_bytes_before_test)) print("server_read_bytes_after_test: {}".format( server_read_bytes_after_test)) if server_read_bytes_after_test == 0: # Not all instance types support RDMA read. # return here for those that does not support return # fi_rdm_tagged_bw is a uni-diretional test. client is the sender and server is the receiver. # In a RDMA read based message tranfer protocol, the receiver will submit work request for read server_read_wrs = server_read_wrs_after_test - server_read_wrs_before_test server_read_bytes = server_read_bytes_after_test - server_read_bytes_before_test client_send_bytes = client_send_bytes_after_test - client_send_bytes_before_test # Among the 256 KB of data, 64 KB data will be sent via 8 RUNTREAD RTM packets. # Each packet has a packet header, therefore the total number of send bytes # is slightly larger than 64 K. assert client_send_bytes > 65536 and client_send_bytes < 66560 # The other 192 KB is transfer by RDMA read # for which the server (receiver) will issue 1 read request. assert server_read_wrs == 1 assert server_read_bytes == 196608
def test_rma_bw_range(cmdline_args, operation_type, completion_type, message_size, memory_type): from efa.efa_common import efa_run_client_server_test command = "fi_rma_bw -e rdm" command = command + " -o " + operation_type efa_run_client_server_test(cmdline_args, command, "short", completion_type, memory_type, message_size)