flume hdfs sink 1个sink同时写500个分区如何配置效率高一些？

Question

agent.sinks.k3.channel = c2
agent.sinks.k3.type = hdfs
agent.sinks.k3.hdfs.path = hdfs://10.31.13.198/flume/%{log_name}/log_date=%{log_date}/server_id=%{server}/
agent.sinks.k3.hdfs.fileType = SequenceFile
agent.sinks.k3.hdfs.codeC = gzip
agent.sinks.k3.hdfs.writeFormat = Text
agent.sinks.k3.hdfs.rollSize = 134217728
agent.sinks.k3.hdfs.rollCount = 0
agent.sinks.k3.hdfs.threadsPoolSize = 500
agent.sinks.k3.hdfs.rollTimerPoolSize = 50
agent.sinks.k3.hdfs.rollInterval = 1800
agent.sinks.k3.hdfs.batchSize = 2000
agent.sinks.k3.hdfs.minBlockReplicas = 1

这是我的配置，希望能提一些建议