Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Flume >> mail # user >> Why flume cannot take all bandwidth.


Copy link to this message
-
Why flume cannot take all bandwidth.
Hello guys:
  I am now doing flume-ng performance test in EC2 instance. And I have a
tier-2 framework, avro client post a 1G file to avro source, and then the
file is store to    HDFS by hdfsSink.
  I wondered why this take about 33ms, network, cpu, memory both have no
pressure. In theory my network can work at 100MB/s, but flume only take
about  60MB/s.
  How can I resolve this problem?
  Thanks a lot. Below is my configure and my test result

# SOURCE
a1.sources = r1
a1.sinks =  k1
a1.channels = c1

a1.sources.r1.type = avro
a1.sources.r1.bind = 10.0.2.13
a1.sources.r1.port = 9876
a1.sources.r1.threads = 10

# SINK (HDFS)
a1.sinks.k1.type = hdfs
a1.sinks.k1.channel = c1
a1.sinks.k1.hdfs.filePrefix = packet
a1.sinks.k1.hdfs.batchSize= 5000
a1.sinks.k1.hdfs.fileSuffix = .snappy
a1.sinks.k1.hdfs.codeC = snappy
a1.sinks.k1.hdfs.fileType = CompressedStream
a1.sinks.k1.hdfs.rollCount = 0
a1.sinks.k1.hdfs.rollSize = 500000000
a1.sinks.k1.hdfs.rollInterval = 0
a1.sinks.k1.hdfs.path = ....

# INTERCEPTORS (TIMESTAMP FOR HDFS PATH)
a1.sources.r1.interceptors = i1 i2
a1.sources.r1.interceptors.i1.type = timestamp
a1.sources.r1.interceptors.i2.type = host
a1.sources.r1.interceptors.i2.preserveExisting = false
a1.sources.r1.interceptors.i2.hostHeader = test-1

# CHANNEL (MEM),take max 1g memory
a1.channels.c1.type = memory
a1.channels.c1.capacity = 10000
a1.channels.c1.transactionCapacity = 5000
a1.channels.c1.byteCapacity = 1000000000
## bind the source and sink to the channel
a1.sources.r1.channels = c1
a1.sinks.k1.channel = c1

my client command: time flume-ng avro-client -c /etc/flume-ng/conf -P
/tmp/rpcProps -F /tmp/flume/tmp/test-1.tmp -Xmx1024m -Xms1024m -Xmn800m
-Xss512k
command result:real 0m33.061s user 0m13.689s sys 0m5.504s