advancedxy commented on code in PR #1650: URL: https://github.com/apache/incubator-uniffle/pull/1650#discussion_r1567522771
########## 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: > 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. It seems like that we have different perspectives on the goal of the Uniffle project. Yes, currently Uniffle's biggest advantage over Vanilla Spark is that it can handle large enough shuffles robustly, and we are prioritizing stability over speed. But ultimately, we want to make Uniffle general enough for almost every case and faster or at least as fast compared with vanilla spark for most cases. That's why we are adding tiered storage support and the rust implementation for potential speedups. Anyway, I think this discussion is slightly off the topic for this PR. Just sharing my perspective of the Uniffle project. -- 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]
