RFR: 8305744: (ch) InputStream returned by Channels.newInputStream should have fast path for ReadbleByteChannel/WritableByteChannel [v3]

[Brian Burkhalter]

On Sun, 9 Apr 2023 15:08:48 GMT, Markus KARG <duke at openjdk.org> wrote:

> Sandbox.ChannelInputStream_transferTo  thrpt   25  3,794 ┬▒ 0,095  ops/s -- incl. optimization (i. e. using ByteBuffer)
> Sandbox.ChannelInputStream_transferTo  thrpt   25  3,703 ┬▒ 0,098  ops/s -- excl. optimization (i. e. using byte[])

If these results are obtained using the default settings, then if I am not mistaken, the following represent the 99.9% confidence intervals:

[3.699, 3.889] incl. optimization
[3.605, 3.801] excl. optimization

As these intervals overlap, it is not really valid in a statistical sense to definitively state that there is an improvement in throughput. In general to have a reasonable level of certainty that there is in fact improvement, the intervals should not overlap at all and the widths of the intervals should be small compared to the magnitudes of the values being compared. The widths of the intervals are small enough here, but given that the intervals overlap by more than one third one must question the significance of the change.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/13387#issuecomment-1511937541