Uploaded image for project: 'Apache Ozone'
  1. Apache Ozone
  2. HDDS-9536

Datanode perf: Copying (heap) buffers is costly

    XMLWordPrintableJSON

Details

    Description

      Today, datanodes don't use direct buffers for WriteChunk data. When the chunks are written to disk, NIO converts those buffers to direct ones and the conversion seems to be very costly (please see attached datanode-on-write2.html).

      Chunk data proto should be serialized from network/ratis using (pooled) direct buffers. That would avoid lots of extra costs not only from buffer copying but also from GCing the immediate buffers.

      Attachments

        1. datanode-on-write2.html
          353 kB
          Duong
        2. datanode-read-ratis.html
          171 kB
          Duong
        3. datanode-read-ratis-after-mapped-buffer.html
          954 kB
          Duong
        4. Screenshot 2023-10-25 at 8.44.16 AM.png
          538 kB
          Duong

        Issue Links

          is fixed by

          Improvement - An improvement or enhancement to an existing feature or task. HDDS-7117 Consider reading chunk files using MappedByteBuffer

          • Major - Major loss of function.
          • Resolved
          is related to

          Sub-task - The sub-task of the issue RATIS-1926 Support Zero-Copy in loading log entries from local files

          • Major - Major loss of function.
          • Open

          Sub-task - The sub-task of the issue RATIS-1925 Support Zero-Copy in GrpcClientProtocolService

          • Major - Major loss of function.
          • Resolved

          Improvement - An improvement or enhancement to an existing feature or task. HDDS-7188 Read chunk files using netty ChunkedNioFile.

          • Major - Major loss of function.
          • Open

          Improvement - An improvement or enhancement to an existing feature or task. HDDS-7117 Consider reading chunk files using MappedByteBuffer

          • Major - Major loss of function.
          • Resolved

          Improvement - An improvement or enhancement to an existing feature or task. HDDS-9611 Avoid unnecessary buffer conversion

          • Major - Major loss of function.
          • Resolved
          requires

          Improvement - An improvement or enhancement to an existing feature or task. HDDS-9936 Create a DispatcherContext for ReadChunk and GetSmallFile

          • Major - Major loss of function.
          • Open
          split to

          Improvement - An improvement or enhancement to an existing feature or task. HDDS-9843 Ozone client high memory (heap) utilization

          • Major - Major loss of function.
          • Resolved
          links to

          Web Link GitHub Pull Request #5497

          Activity

            People

              szetszwo Tsz-wo Sze
              duongnguyen Duong
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: