Chung-En Lee created HDDS-15734:
-----------------------------------
Summary: Optimize `readVectored` by implementing
`ByteBufferPositionedReadable` in `BlockInputStream` to reduce lock contention
Key: HDDS-15734
URL: https://issues.apache.org/jira/browse/HDDS-15734
Project: Apache Ozone
Issue Type: Sub-task
Reporter: Chung-En Lee
Currently, when performing parallel read operations via {{{}readVectored{}}},
we are experiencing significant lock contention. This bottleneck happens
because {{BlockInputStream}} and {{ChunkInputStream}} do not implement
{{{}ByteBufferPositionedReadable{}}}. As a result, concurrent threads end up
competing for the same underlying stream locks during positioned read
operations.
h4. The goal of this ticket is to implement ByteBufferPositionedReadable for
both BlockInputStream and ChunkInputStream. By enabling thread-safe,
position-based reads directly into ByteBuffer, we can bypass the
synchronization overhead and greatly improve the performance of concurrent
readVectored calls.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
