Github user liyezhang556520 commented on the pull request: https://github.com/apache/spark/pull/12083#issuecomment-204207886 Hi @vanzin , the memory copy place is given out by @zsxwing , the call stack is as follows: ``` at java.nio.Bits.copyFromArray(Bits.java:754) at java.nio.DirectByteBuffer.put(DirectByteBuffer.java:371) at java.nio.DirectByteBuffer.put(DirectByteBuffer.java:342) at sun.nio.ch.IOUtil.write(IOUtil.java:60) at sun.nio.ch.SocketChannelImpl.write(SocketChannelImpl.java:466) - locked <0x00007f8a8a28d400> (a java.lang.Object) at org.apache.spark.network.protocol.MessageWithHeader.copyByteBuf(MessageWithHeader.java:131) at org.apache.spark.network.protocol.MessageWithHeader.transferTo(MessageWithHeader.java:114) ``` The whole buffer copy is in line http://www.grepcode.com/file/repository.grepcode.com/java/root/jdk/openjdk/7u40-b43/sun/nio/ch/IOUtil.java#60, but the buffer cannot be totally written if its side more than the available underlying buffer. Which is in line http://www.grepcode.com/file/repository.grepcode.com/java/root/jdk/openjdk/7u40-b43/sun/nio/ch/IOUtil.java#65. So each time we will make a copy of the input `ByteBuf`, and write only a part of it if the input size is big relatively. This results in multiply copies of the input `ByteBuf` that is not necessary. The method of handling the issue in this PR is the same as that in Hadoop, please refer to https://github.com/apache/hadoop/blob/trunk/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/ipc/Server.java#L2957
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org