Skip to content

Latest commit

 

History

History
118 lines (93 loc) · 4.47 KB

gcp.md

File metadata and controls

118 lines (93 loc) · 4.47 KB

Get Started with XGBoost4J-Spark on Google Cloud

This is a getting started guide to XGBoost4J-Spark on Google Cloud. At the end of this guide, the reader will be able to run a sample Apache Spark application that runs on NVIDIA GPUs on Google Cloud.

Install & Cluster Spin up

Then, use the steps from this blog to create a RAPIDS cluster on Dataproc. Specifically, run through the steps in the Install Google Cloud SDK and the Create a Dataproc cluster sections.

Here’s an example command:

CLUSTER_NAME=raghav-cluster
ZONE=us-east1-c
STORAGE_BUCKET=rapidsai-test-1
NUM_GPUS_IN_MASTER=4
NUM_GPUS_IN_WORKER=4
NUM_WORKERS=2
DATAPROC_BUCKET=dataproc-initialization-actions

gcloud beta dataproc clusters create $CLUSTER_NAME  \
--zone $ZONE \
--master-accelerator type=nvidia-tesla-t4,count=$NUM_GPUS_IN_MASTER \
--master-machine-type n1-standard-32 \
--worker-accelerator type=nvidia-tesla-t4,count=$NUM_GPUS_IN_WORKER \
--worker-machine-type n1-standard-32 \
--num-workers $NUM_WORKERS \
--bucket $STORAGE_BUCKET \
--metadata "JUPYTER_PORT=8888" \
--initialization-actions gs://$DATAPROC_BUCKET/rapids/rapids.sh \
--optional-components=ANACONDA,JUPYTER \
--enable-component-gateway

This cluster should now have NVIDIA drivers and the CUDA run time.

Submitting Jobs

Set up the jar files and data on a root folder in a GCP storage bucket like below.

You can either drag and drop files from the GCP storage browser, or use the gsutil cp to do this from the command line.

GCP storage bucket

Submit Spark Job on GPUs

Use the following command to submit spark jobs on this GPU cluster.

CLUSTER_NAME=raghav-cluster
ROOT_FOLDER=gs://rapidsai-test-1/spark
MAIN_CLASS=ai.rapids.spark.examples.agaricus.GPUMain
RAPIDS_JARS=$ROOT_FOLDER/jars/sample_xgboost_apps-0.1.4.jar,$ROOT_FOLDER/jars/cudf-0.9-cuda10.jar,$ROOT_FOLDER/jars/xgboost4j_2.11-1.0.0-Beta.jar,$ROOT_FOLDER/jars/xgboost4j-spark_2.11-1.0.0-Beta.jar
TRAIN_DATA_PATH=$ROOT_FOLDER/data/agaricus/csv/train
TEST_DATA_PATH=$ROOT_FOLDER/data/agaricus/csv/test
MODEL_PATH=$ROOT_FOLDER/models

gcloud beta dataproc jobs submit spark \
--cluster=$CLUSTER_NAME \
--class=$MAIN_CLASS \
--jars=$RAPIDS_JARS \
--properties=spark.executor.memory=32G,spark.executorEnv.LD_LIBRARY_PATH=/opt/conda/anaconda/envs/RAPIDS/lib/ \
-- \
-format=csv \
-numRound=100 \
-mode=train \
-trainDataPath=$TRAIN_DATA_PATH \
-trainDataPath=$TEST_DATA_PATH \
-modelPath=$MODEL_PATH \
-overwrite=true

Here's a quick explanation of key parameters:

  • CLUSTER_NAME: name of the cluster you had created in the first step
  • ROOT_FOLDER: Storage bucket on GCP where you organize your data and jar files
  • MAIN_CLASS: Class containing the main method to run
  • RAPIDS_JARS: All the jar files you file need. This includes the RAPIDS jars as well as the one for the application you are submitting. properties: Use this to specify Spark properties. The command above includes the ones you likely need.

You can check out the full documentation of this api here.

Submit Spark Job on CPUs

Submitting a CPU job on this cluster is very similar. Below's an example command that runs the same Agaricus application on CPUs:

CLUSTER_NAME=raghav-cluster
ROOT_FOLDER=gs://rapidsai-test-1/spark
MAIN_CLASS=ai.rapids.spark.examples.agaricus.CPUMain
RAPIDS_JARS=$ROOT_FOLDER/jars/sample_xgboost_apps-0.1.4.jar,$ROOT_FOLDER/jars/cudf-0.9-cuda10.jar,$ROOT_FOLDER/jars/xgboost4j_2.11-1.0.0-Beta.jar,$ROOT_FOLDER/jars/xgboost4j-spark_2.11-1.0.0-Beta.jar
TRAIN_DATA_PATH=$ROOT_FOLDER/data/agaricus/csv/train
TEST_DATA_PATH=$ROOT_FOLDER/data/agaricus/csv/test
MODEL_PATH=$ROOT_FOLDER/models

gcloud beta dataproc jobs submit spark \
--cluster=$CLUSTER_NAME \
--class=$MAIN_CLASS \
--jars=$RAPIDS_JARS \
--properties=spark.executor.memory=32G \
-- \
-format=csv \
-numRound=100 \
-mode=train \
-trainDataPath=$TRAIN_DATA_PATH \
-trainDataPath=$TEST_DATA_PATH \
-modelPath=$MODEL_PATH \
-overwrite=true \
-numWorkers=12 \
-treeMethod=hist

Clean Up

When you're done working on this cluster, don't forget to delete the cluster, using the following command (replacing the highlighted cluster name with yours):

gcloud dataproc clusters delete raghav-cluster