[
https://issues.apache.org/jira/browse/HBASE-5945?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stack updated HBASE-5945:
-------------------------
Attachment: 5945v2.txt
Forward port Todd's old idea of saving copies by passing array of ByteBuffers
to Channel rather than compose response in a buffer first before putting it on
the wire. Revives his BufferChain doohickey.
Here's some comment on the changes.
M hbase-client/src/main/java/org/apache/hadoop/hbase/ipc/IPCUtil.java
Added getDelimitedMessageAsByteBuffer(final Message m)
Removed no longer used write method and getDelimitedMessageBytes.
M
hbase-hadoop-compat/src/main/java/org/apache/hadoop/hbase/ipc/MetricsHBaseServerSource.java
Changed bytes sent size from int to long (it is long in implementation, just
not in interfaces)
A hbase-server/src/main/java/org/apache/hadoop/hbase/ipc/BufferChain.java
Class to carry an array of ByteBuffers and that can write the byte buffers in
order
out to a GatheringByteChannel in chunks.
M hbase-server/src/main/java/org/apache/hadoop/hbase/ipc/RpcServer.java
Refactored response sending to use BufferChain. We save on a composition of
total
response in memory. Instead we now pass the parts directly to the socket
channel.
Stuff is a bit ugly in the sasl handling because have to compose the message
in memory completely so can do the sasl wrapping; this undoes any savings but
I believe we are still ahead because we removed the serializing of the resulting
sasl array to an output stream backed by a new buffer.
> Reduce buffer copies in IPC server response path
> ------------------------------------------------
>
> Key: HBASE-5945
> URL: https://issues.apache.org/jira/browse/HBASE-5945
> Project: HBase
> Issue Type: Improvement
> Components: IPC/RPC
> Affects Versions: 0.95.2
> Reporter: Todd Lipcon
> Assignee: stack
> Fix For: 0.96.1
>
> Attachments: 5945-in-progress.2.1.patch, 5945-in-progress.2.patch,
> 5945-in-progress.patch, 5945v2.txt, buffer-copies.txt, even-fewer-copies.txt,
> hbase-5495.txt
>
>
> The new PB code is sloppy with buffers and makes several needless copies.
> This increases GC time a lot. A few simple changes can cut this back down.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
