[
https://issues.apache.org/jira/browse/IGNITE-7097?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16292718#comment-16292718
]
Oleg Ignatenko edited comment on IGNITE-7097 at 12/15/17 4:01 PM:
------------------------------------------------------------------
Upon further study the root cause of the issue turned out that the
implementation in this matrix type is very slow, much slower than in all other
classes used in matrix mul benchmarks. Based on that the way to resolve this
was chosen as follows: this benchmark runs on matrices of smaller size than
other, plus it is complemented with a benchmark that uses regular size but also
applies a trivial optimization by delegating multiplication to
{{SparseBlockDistributedMatrix}}.
For the sake of completeness I also briefly discussed with Yury an option to
somehow optimize multiplication in this type of matrix. As of now it doesn't
look worth the effort. This is primarily because we already have a properly
optimized block version of sparse distributed matrix, so is unclear what
tangible benefits can be gained by such an optimization.
was (Author: oignatenko):
On a closer inspection the root cause of the issue turned out that the
implementation in this matrix type is very slow, much slower than in all other
classes used in matrix mul benchmarks. Based on that the way to resolve this
was chosen as follows: this benchmark runs on matrices of smaller size than
other, plus it is complemented with a benchmark that uses regular size but also
applies a trivial optimization by delegating multiplication to
{{SparseBlockDistributedMatrix}}.
For the sake of completeness I also briefly discussed with Yury an option to
somehow optimize multiplication in this type of matrix. As of now it doesn't
look worth the effort. This is primarily because we already have a properly
optimized block version of sparse distributed matrix, so is unclear what
tangible benefits can be gained by such an optimization.
> performance measurement for SparseDistributedMatrix multiplication
> ------------------------------------------------------------------
>
> Key: IGNITE-7097
> URL: https://issues.apache.org/jira/browse/IGNITE-7097
> Project: Ignite
> Issue Type: Task
> Components: ml, yardstick
> Reporter: Oleg Ignatenko
> Assignee: Oleg Ignatenko
> Fix For: 2.4
>
>
> We want to start tracking our performance to avoid performance degradation.
> Also we need some performance comparison with other ml libs.
> Initial draft for this benchmark was made per IGNITE-6123 (class
> {{IgniteSparseDistributedMatrixMulBenchmark}}) but it currently hangs so it
> is excluded. Find a way to do it right.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
