rickyma commented on code in PR #1650: URL: https://github.com/apache/incubator-uniffle/pull/1650#discussion_r1567421959
########## docs/benchmark_netty.md: ########## @@ -0,0 +1,132 @@ +<!-- + ~ Licensed to the Apache Software Foundation (ASF) under one or more + ~ contributor license agreements. See the NOTICE file distributed with + ~ this work for additional information regarding copyright ownership. + ~ The ASF licenses this file to You under the Apache License, Version 2.0 + ~ (the "License"); you may not use this file except in compliance with + ~ the License. You may obtain a copy of the License at + ~ + ~ http://www.apache.org/licenses/LICENSE-2.0 + ~ + ~ Unless required by applicable law or agreed to in writing, software + ~ distributed under the License is distributed on an "AS IS" BASIS, + ~ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + ~ See the License for the specific language governing permissions and + ~ limitations under the License. + --> + +## Environment + +### Software + +Uniffle 0.9.0, Hadoop 2.8.5, Spark 3.3.1 + +### Hardware + +#### Uniffle Cluster + +| Cluster Type | Memory | CPU Cores | Disk Configuration for Every Shuffle Server | Max IO Read/Write Speed | Quantity | Network Bandwidth | +|--------------|--------|-----------|---------------------------------------------|-------------------------|---------------------------------------|-------------------| +| HDD | 250G | 96 | 10 * 4T HDD | 150MB/s | 2 * Coordinator + 10 * Shuffle Server | 25GB/s | +| SSD | 250G | 96 | 1 * 6T NVME | 3GB/s | 2 * Coordinator + 10 * Shuffle Server | 25GB/s | + +#### Hadoop Yarn Cluster + +2 * ResourceManager + 750 * NodeManager, every machine 12 * 4T HDD + +## Configuration + +Spark's configuration: + + ```` + spark.speculation false + spark.executor.instances 1400 + spark.executor.cores 2 + spark.executor.memory 20g + spark.executor.memoryOverhead 1024 + spark.shuffle.manager org.apache.spark.shuffle.RssShuffleManager + spark.sql.shuffle.partitions 20000 Review Comment: When giving a more comprehensive report, it can be done this way, but it will take some time. If we are only looking from the perspective of Uniffle, there is no need to spend time comparing scenarios where Vanilla Spark has more advantages (such as smaller task concurrency, smaller shuffle data size), because Uniffle is not meant to be faster than Vanilla Spark. As long as Uniffle can be more stable than Vanilla Spark in any scenario during shuffling, and can handle larger task concurrency and larger shuffle size, I think that's enough. What we pursue is stability. You cannot predict how many partitions the user's Job will have, so in my case, Vanilla Spark cannot run successfully, but Uniffle can. That's enough. We don't necessarily need to list cases where Vanilla Spark can successfully run a job with fewer partitions, as it doesn't have much significance for Uniffle. -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
