CDH 5.1 was installed using Cloudera Manager 5.1.1. There are 3 master nodes and 10 data nodes in the cluster. Each nodes has 32GB RAM. After the Hive Metastore Server is restarted, it only lasts for a couple of minutes. The resident memory usage reaches 1GB (default setting) before the process crashes. After increasing java heap size of HMS to 6GB, it lasts about 25 minutes and usage of memory was up to 5.8GB before the process crashes. Oracle database is used as the data store. There are aobut 11 to 12 sessions connecting to database when HMS is running. Hive is not functional because HMS is not up. Any advices would be appreciated.