when i use benchmark to test performance?
distributed_replicated is worse than parameter_server. anyone meet the same question?
1worker and 1 ps.
my start code is as below:
python2 /root/benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py \
--server_protocol=grpc+verbs \
--data_dir=/root/img/ --data_name=imagenet \
--model=vgg16 --variable_update=distributed_replicated \
--batch_size=64 --num_batches=50 --num_warmup_batches=10 \
--local_parameter_device=gpu --num_gpus=1 \
--job_name=ps --task_index=0 \
--ps_hosts=172.168.30.25:10011 \
--worker_hosts=172.168.30.26:50012 &
when i use benchmark to test performance? distributed_replicated is worse than parameter_server. anyone meet the same question? 1worker and 1 ps. my start code is as below: python2 /root/benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py \ --server_protocol=grpc+verbs \ --data_dir=/root/img/ --data_name=imagenet \ --model=vgg16 --variable_update=distributed_replicated \ --batch_size=64 --num_batches=50 --num_warmup_batches=10 \ --local_parameter_device=gpu --num_gpus=1 \ --job_name=ps --task_index=0 \ --ps_hosts=172.168.30.25:10011 \ --worker_hosts=172.168.30.26:50012 &