Moovlin opened a new pull request #29068: URL: https://github.com/apache/spark/pull/29068
This was a dead simple change that I lightly tested to determine if there was actually a performance increase. Turns out, yes there is (at least locally). <!-- Thanks for sending a pull request! Here are some tips for you: 1. If this is your first time, please read our contributor guidelines: https://spark.apache.org/contributing.html 2. Ensure you have added or run the appropriate tests for your PR: https://spark.apache.org/developer-tools.html 3. If the PR is unfinished, add '[WIP]' in your PR title, e.g., '[WIP][SPARK-XXXX] Your PR title ...'. 4. Be sure to keep the PR description updated to reflect all changes. 5. Please write your PR title to summarize what this PR proposes. 6. If possible, provide a concise example to reproduce the issue for a faster review. 7. If you want to add a new configuration, please read the guideline first for naming configurations in 'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'. --> ### What changes were proposed in this pull request? <!-- Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. See the examples below. 1. If you refactor some codes with changing classes, showing the class hierarchy will help reviewers. 2. If you fix some SQL features, you can provide some references of other DBMSes. 3. If there is design documentation, please add the link. 4. If there is a discussion in the mailing list, please add the link. --> As per the issue SPARK-27892, the saving & loading of files was being done sequentially in the SharedReadWrite object which would clearly make loading large models quite slow. The enhancement simply creates a new ParArray & converts the existing stage array into a parallel array by calling ".par". When passed into the foreach / mapper phases it is automatically parallelized. ### Why are the changes needed? <!-- Please clarify why the changes are needed. For instance, 1. If you propose a new API, clarify the use case for a new API. 2. If you fix a bug, you can clarify why it is a bug. --> This only serves to speed up load times of pipelines with many stages. ### Does this PR introduce _any_ user-facing change? <!-- Note that it means *any* user-facing change including all aspects such as the documentation fix. If yes, please clarify the previous behavior and the change this PR proposes - provide the console output, description and/or an example to show the behavior difference if possible. If possible, please also clarify if this is a user-facing change compared to the released Spark versions or within the unreleased branches such as master. If no, write 'No'. --> No. ### How was this patch tested? <!-- If tests were added, say they were added here. Please make sure to add some test cases that check the changes thoroughly including negative and positive cases if possible. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> I tested it by using the example provided in SPARK-27892. I also added a load step to verify that worked correctly. `import org.apache.spark.ml._ def time[R](block : => R): R = { val start = System.nanoTime() val result = block val end = System.nanoTime() println("elap time: " + (end - start) + "ns") result } import org.apache.spark.ml.feature.VectorAssembler val outputPath = "boopcity" val stages = (1 to 100) map { i => new VectorAssembler().setInputCols(Array("input")).setOutputCol("o" + i)} val p = new Pipeline().setStages(stages.toArray) val data = Seq(1, 1, 1) toDF "input" val pm = p.fit(data) val result = time { pm.write.overwrite().save(outputPath) } val sameModel = time{ PipelineModel.load(outputPath) }` I compared the execution time within Scala by running 4 runs for both the sequential and parallel version. The raw "data" para: save: 11294731300ns 11211858600ns 7047186600ns 6382136700ns mean: 8983978300ns load: 6909773600ns 15572318900ns 5430449700ns 4499734100ns mean: 8103069075ns seq: save: 43879181600ns 27545280700ns 24504610300ns 23568721400ns mean: load: 27151395000ns 20363685700ns 15923967900ns 15889010200ns para: save: 8983978300ns load: 8103069075ns seq: save: 20029586625ns load: 19832014700ns Decreasing times are likely contributed to the data simply being cached on faster portions of the SSD in my machine so the average isn't super meaningful but consistently we see that the parallel version is far faster than sequential. Additionally, I ran the entire mllib testsuite on the changes with all test passing. Testing on a cluster is likely worth doing but I don't have the resources to do that at the moment. There is no way to add a unit test to do this without adding an additional function which doesn't isn't done in parallel which seems like a waste of time & effort to add. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org