The document discusses optimizations made to Spark SQL performance when working with Parquet files at ByteDance. It describes how Spark originally reads Parquet files and identifies two main areas for optimization: Parquet filter pushdown and the Parquet reader. For filter pushdown, sorting columns improved statistics and reduced data reads by 30%. For the reader, splitting it to first filter then read other columns prevented loading unnecessary data. These changes improved Spark SQL performance at ByteDance without changing jobs.
Hive Bucketing in Apache Spark with Tejas PatilDatabricks
Bucketing is a partitioning technique that can improve performance in certain data transformations by avoiding data shuffling and sorting. The general idea of bucketing is to partition, and optionally sort, the data based on a subset of columns while it is written out (a one-time cost), while making successive reads of the data more performant for downstream jobs if the SQL operators can make use of this property. Bucketing can enable faster joins (i.e. single stage sort merge join), the ability to short circuit in FILTER operation if the file is pre-sorted over the column in a filter predicate, and it supports quick data sampling.
In this session, you’ll learn how bucketing is implemented in both Hive and Spark. In particular, Patil will describe the changes in the Catalyst optimizer that enable these optimizations in Spark for various bucketing scenarios. Facebook’s performance tests have shown bucketing to improve Spark performance from 3-5x faster when the optimization is enabled. Many tables at Facebook are sorted and bucketed, and migrating these workloads to Spark have resulted in a 2-3x savings when compared to Hive. You’ll also hear about real-world applications of bucketing, like loading of cumulative tables with daily delta, and the characteristics that can help identify suitable candidate jobs that can benefit from bucketing.
This document discusses Spark shuffle, which is an expensive operation that involves data partitioning, serialization/deserialization, compression, and disk I/O. It provides an overview of how shuffle works in Spark and the history of optimizations like sort-based shuffle and an external shuffle service. Key concepts discussed include shuffle writers, readers, and the pluggable block transfer service that handles data transfer. The document also covers shuffle-related configuration options and potential future work.
Spark Shuffle Deep Dive (Explained In Depth) - How Shuffle Works in SparkBo Yang
The slides explain how shuffle works in Spark and help people understand more details about Spark internal. It shows how the major classes are implemented, including: ShuffleManager (SortShuffleManager), ShuffleWriter (SortShuffleWriter, BypassMergeSortShuffleWriter, UnsafeShuffleWriter), ShuffleReader (BlockStoreShuffleReader).
Deep Dive: Memory Management in Apache SparkDatabricks
Memory management is at the heart of any data-intensive system. Spark, in particular, must arbitrate memory allocation between two main use cases: buffering intermediate data for processing (execution) and caching user data (storage). This talk will take a deep dive through the memory management designs adopted in Spark since its inception and discuss their performance and usability implications for the end user.
Top 5 Mistakes When Writing Spark ApplicationsSpark Summit
This document discusses 5 common mistakes when writing Spark applications:
1) Improperly sizing executors by not considering cores, memory, and overhead. The optimal configuration depends on the workload and cluster resources.
2) Applications failing due to shuffle blocks exceeding 2GB size limit. Increasing the number of partitions helps address this.
3) Jobs running slowly due to data skew in joins and shuffles. Techniques like salting keys can help address skew.
4) Not properly managing the DAG to avoid shuffles and bring work to the data. Using ReduceByKey over GroupByKey and TreeReduce over Reduce when possible.
5) Classpath conflicts arising from mismatched library versions, which can be addressed using sh
Common Strategies for Improving Performance on Your Delta LakehouseDatabricks
The Delta Architecture pattern has made the lives of data engineers much simpler, but what about improving query performance for data analysts? What are some common places to look at for tuning query performance? In this session we will cover some common techniques to apply to our delta tables to make them perform better for data analysts queries. We will look at a few examples of how you can analyze a query, and determine what to focus on to deliver better performance results.
How We Optimize Spark SQL Jobs With parallel and sync IODatabricks
Although NVMe has been more and more popular these years, a large amount of HDD are still widely used in super-large scale big data clusters. In a EB-level data platform, IO(including decompression and decode) cost contributes a large proportion of Spark jobs’ cost. In another word, IO operation is worth optimizing.
In ByteDancen, we do a series of IO optimization to improve performance, including parallel read and asynchronized shuffle. Firstly we implement file level parallel read to improve performance when there are a lot of small files. Secondly, we design row group level parallel read to accelerate queries for big-file scenario. Thirdly, implement asynchronized spill to improve job peformance. Besides, we design parquet column family, which will split a table into a few column families and different column family will be in different Parquets files. Different column family can be read in parallel, so the read performance is much higher than the existing approach. In our practice, the end to end performance is improved by 5% to 30%
In this talk, I will illustrate how we implement these features and how they accelerate Apache Spark jobs.
Adaptive Query Execution: Speeding Up Spark SQL at RuntimeDatabricks
Over the years, there has been extensive and continuous effort on improving Spark SQL’s query optimizer and planner, in order to generate high quality query execution plans. One of the biggest improvements is the cost-based optimization framework that collects and leverages a variety of data statistics (e.g., row count, number of distinct values, NULL values, max/min values, etc.) to help Spark make better decisions in picking the most optimal query plan.
A Thorough Comparison of Delta Lake, Iceberg and HudiDatabricks
Recently, a set of modern table formats such as Delta Lake, Hudi, Iceberg spring out. Along with Hive Metastore these table formats are trying to solve problems that stand in traditional data lake for a long time with their declared features like ACID, schema evolution, upsert, time travel, incremental consumption etc.
Deep Dive into the New Features of Apache Spark 3.0Databricks
Continuing with the objectives to make Spark faster, easier, and smarter, Apache Spark 3.0 extends its scope with more than 3000 resolved JIRAs. We will talk about the exciting new developments in the Spark 3.0 as well as some other major initiatives that are coming in the future.
From Query Plan to Query Performance: Supercharging your Apache Spark Queries...Databricks
The SQL tab in the Spark UI provides a lot of information for analysing your spark queries, ranging from the query plan, to all associated statistics. However, many new Spark practitioners get overwhelmed by the information presented, and have trouble using it to their benefit. In this talk we want to give a gentle introduction to how to read this SQL tab. We will first go over all the common spark operations, such as scans, projects, filter, aggregations and joins; and how they relate to the Spark code written. In the second part of the talk we will show how to read the associated statistics to pinpoint performance bottlenecks.
A Deep Dive into Query Execution Engine of Spark SQLDatabricks
Spark SQL enables Spark to perform efficient and fault-tolerant relational query processing with analytics database technologies. The relational queries are compiled to the executable physical plans consisting of transformations and actions on RDDs with the generated Java code. The code is compiled to Java bytecode, executed at runtime by JVM and optimized by JIT to native machine code at runtime. This talk will take a deep dive into Spark SQL execution engine. The talk includes pipelined execution, whole-stage code generation, UDF execution, memory management, vectorized readers, lineage based RDD transformation and action.
Tuning Apache Spark for Large-Scale Workloads Gaoxiang Liu and Sital KediaDatabricks
Apache Spark is a fast and flexible compute engine for a variety of diverse workloads. Optimizing performance for different applications often requires an understanding of Spark internals and can be challenging for Spark application developers. In this session, learn how Facebook tunes Spark to run large-scale workloads reliably and efficiently. The speakers will begin by explaining the various tools and techniques they use to discover performance bottlenecks in Spark jobs. Next, you’ll hear about important configuration parameters and their experiments tuning these parameters on large-scale production workload. You’ll also learn about Facebook’s new efforts towards automatically tuning several important configurations based on nature of the workload. The speakers will conclude by sharing their results with automatic tuning and future directions for the project.ing several important configurations based on nature of the workload. We will conclude by sharing our result with automatic tuning and future directions for the project.
Join operations in Apache Spark is often the biggest source of performance problems and even full-blown exceptions in Spark. After this talk, you will understand the two most basic methods Spark employs for joining DataFrames – to the level of detail of how Spark distributes the data within the cluster. You’ll also find out how to work out common errors and even handle the trickiest corner cases we’ve encountered! After this talk, you should be able to write performance joins in Spark SQL that scale and are zippy fast!
This session will cover different ways of joining tables in Apache Spark.
Speaker: Vida Ha
This talk was originally presented at Spark Summit East 2017.
Building a SIMD Supported Vectorized Native Engine for Spark SQLDatabricks
Spark SQL works very well with structured row-based data. Vectorized reader and writer for parquet/orc can make I/O much faster. It also used WholeStageCodeGen to improve the performance by Java JIT code. However Java JIT is usually not working very well on utilizing latest SIMD instructions under complicated queries. Apache Arrow provides columnar in-memory layout and SIMD optimized kernels as well as a LLVM based SQL engine Gandiva. These native based libraries can accelerate Spark SQL by reduce the CPU usage for both I/O and execution.
Apache Spark Data Source V2 with Wenchen Fan and Gengliang WangDatabricks
As a general computing engine, Spark can process data from various data management/storage systems, including HDFS, Hive, Cassandra and Kafka. For flexibility and high throughput, Spark defines the Data Source API, which is an abstraction of the storage layer. The Data Source API has two requirements.
1) Generality: support reading/writing most data management/storage systems.
2) Flexibility: customize and optimize the read and write paths for different systems based on their capabilities.
Data Source API V2 is one of the most important features coming with Spark 2.3. This talk will dive into the design and implementation of Data Source API V2, with comparison to the Data Source API V1. We also demonstrate how to implement a file-based data source using the Data Source API V2 for showing its generality and flexibility.
Top 5 Mistakes to Avoid When Writing Apache Spark ApplicationsCloudera, Inc.
The document discusses 5 common mistakes people make when writing Spark applications:
1) Not properly sizing executors for memory and cores.
2) Having shuffle blocks larger than 2GB which can cause jobs to fail.
3) Not addressing data skew which can cause joins and shuffles to be very slow.
4) Not properly managing the DAG to minimize shuffles and stages.
5) Classpath conflicts from mismatched dependencies causing errors.
Memory management is at the heart of any data-intensive system. Spark, in particular, must arbitrate memory allocation between two main use cases: buffering intermediate data for processing (execution) and caching user data (storage). This talk will take a deep dive through the memory management designs adopted in Spark since its inception and discuss their performance and usability implications for the end user.
This is the presentation I made on JavaDay Kiev 2015 regarding the architecture of Apache Spark. It covers the memory model, the shuffle implementations, data frames and some other high-level staff and can be used as an introduction to Apache Spark
In Spark SQL the physical plan provides the fundamental information about the execution of the query. The objective of this talk is to convey understanding and familiarity of query plans in Spark SQL, and use that knowledge to achieve better performance of Apache Spark queries. We will walk you through the most common operators you might find in the query plan and explain some relevant information that can be useful in order to understand some details about the execution. If you understand the query plan, you can look for the weak spot and try to rewrite the query to achieve a more optimal plan that leads to more efficient execution.
The main content of this talk is based on Spark source code but it will reflect some real-life queries that we run while processing data. We will show some examples of query plans and explain how to interpret them and what information can be taken from them. We will also describe what is happening under the hood when the plan is generated focusing mainly on the phase of physical planning. In general, in this talk we want to share what we have learned from both Spark source code and real-life queries that we run in our daily data processing.
What’s New in the Upcoming Apache Spark 3.0Databricks
Learn about the latest developments in the open-source community with Apache Spark 3.0 and DBR 7.0The upcoming Apache Spark™ 3.0 release brings new capabilities and features to the Spark ecosystem. In this online tech talk from Databricks, we will walk through updates in the Apache Spark 3.0.0-preview2 release as part of our new Databricks Runtime 7.0 Beta, which is now available.
Extreme Apache Spark: how in 3 months we created a pipeline that can process ...Josef A. Habdank
Presentation consists of an amazing bundle of Pro tips and tricks for building an insanely scalable Apache Spark and Spark Streaming based data pipeline.
Presentation consists of 4 parts:
* Quick intro to Spark
* N-billion rows/day system architecture
* Data Warehouse and Messaging
* How to deploy spark so it does not backfire
From HDFS to S3: Migrate Pinterest Apache Spark ClustersDatabricks
The document discusses Pinterest migrating their Apache Spark clusters from HDFS to S3 storage. Some key points:
1) Migrating to S3 provided significantly better performance due to the higher IOPS of modern EC2 instances compared to their older HDFS nodes. Jobs saw 25-35% improvements on average.
2) S3 is eventually consistent while HDFS is strongly consistent, so they implemented the S3Committer to handle output consistency issues during job failures.
3) Metadata operations like file moves were very slow in S3, so they optimized jobs to reduce unnecessary moves using techniques like multipart uploads to S3.
Spark Summit EU 2015: Lessons from 300+ production usersDatabricks
At Databricks, we have a unique view into over a hundred different companies trying out Spark for development and production use-cases, from their support tickets and forum posts. Having seen so many different workflows and applications, some discernible patterns emerge when looking at common performance and scalability issues that our users run into. This talk will discuss some of these common common issues from an engineering and operations perspective, describing solutions and clarifying misconceptions.
Healthcare Claim Reimbursement using Apache SparkDatabricks
The document discusses rewriting a claims reimbursement system using Spark. It describes how Spark provides better performance, scalability and cost savings compared to the previous Oracle-based system. Key points include using Spark for ETL to load data into a Delta Lake data lake, implementing the business logic in a reusable Java library, and seeing significant increases in processing volumes and speeds compared to the prior system. Challenges and tips for adoption are also provided.
(BDT303) Running Spark and Presto on the Netflix Big Data PlatformAmazon Web Services
In this session, we discuss how Spark and Presto complement the Netflix big data platform stack that started with Hadoop, and the use cases that Spark and Presto address. Also, we discuss how we run Spark and Presto on top of the Amazon EMR infrastructure; specifically, how we use Amazon S3 as our data warehouse and how we leverage Amazon EMR as a generic framework for data-processing cluster management.
Running Presto and Spark on the Netflix Big Data PlatformEva Tse
This document summarizes Netflix's big data platform, which uses Presto and Spark on Amazon EMR and S3. Key points:
- Netflix processes over 50 billion hours of streaming per quarter from 65+ million members across over 1000 devices.
- Their data warehouse contains over 25PB stored on S3. They read 10% daily and write 10% of reads.
- They use Presto for interactive queries and Spark for both batch and iterative jobs.
- They have customized Presto and Spark for better performance on S3 and Parquet, and contributed code back to open source projects.
- Their architecture leverages dynamic EMR clusters with Presto and Spark deployed via bootstrap actions for scalability.
This document discusses Netflix's use of Spark on Yarn for ETL workloads. Some key points:
- Netflix runs Spark on Yarn across 3000 EC2 nodes to process large amounts of streaming data from over 100 million daily users.
- Technical challenges included optimizing performance for S3, dynamic resource allocation, and Parquet read/write. Improvements led to up to 18x faster job completion times.
- Production Spark applications include recommender systems that analyze user behavior and personalize content across billions of profiles and titles.
This document discusses Netflix's use of Spark on Yarn for ETL workloads. Some key points:
- Netflix runs Spark on Yarn across 3000 EC2 nodes to process large amounts of streaming data from over 100 million daily users.
- Technical challenges included optimizing performance for S3, dynamic resource allocation, and Parquet read/write. Improvements led to up to 18x faster job completion times.
- Production Spark applications include recommender systems that analyze user behavior and personalize content across billions of profiles and titles.
- Treasure Data is a data analytics platform that unifies raw data from over 100 sources in a scalable and secure manner. It stores data on cloud storage like S3.
- Its storage system called Plazma stores metadata in PostgreSQL and data files in S3. Data is partitioned based on time ranges for efficient querying.
- Presto is used as the distributed query engine. The Treasure Data connector implements metadata, splitting, and data access functions to allow Presto to query data stored in Plazma on S3. This utilizes time partitioning and predicate pushdown for high performance queries.
Operating and Supporting Delta Lake in ProductionDatabricks
The document discusses strategies for optimizing and managing metadata in Delta Lake. It provides an overview of optimize, auto-optimize, and optimize write strategies and how to choose the appropriate strategy based on factors like workload, data size, and cluster resources. It also discusses Delta Lake transaction logs, configurations like log retention duration, and tips for working with Delta Lake metadata.
Apache Spark 3.0: Overview of What’s New and Why CareDatabricks
Spark 3.0 introduces several new features and enhancements to improve performance, usability and compatibility. Key highlights include adaptive query execution which optimizes query plans at runtime based on statistics, dynamic partition pruning to avoid unnecessary data scans, and join hints to influence join strategies. Usability is improved with richer APIs like pandas UDF enhancements and a new structured streaming UI. Compatibility and extensibility is enhanced with Java 11 support, Hive 3.x metastore support and Hadoop 3 support.
Dongwon Kim – A Comparative Performance Evaluation of FlinkFlink Forward
This document provides a summary and analysis of a performance evaluation comparing the big data processing engine Flink to other engines like Spark, Tez, and MapReduce. The key points are:
- Flink completes a 3.2TB TeraSort benchmark faster than Spark, Tez, and MapReduce due to its pipelined execution model which allows more overlap between stages compared to the other engines.
- While Tez and Spark attempt to overlap stages, in practice they do not due to the way tasks are scheduled and launched. MapReduce shows some overlap but is still slower.
- Flink causes fewer disk accesses during shuffling by transferring data directly from memory to memory instead of writing to disk like
A Comparative Performance Evaluation of Apache FlinkDongwon Kim
I compare Apache Flink to Apache Spark, Apache Tez, and MapReduce in Apache Hadoop in terms of performance. I run experiments using two benchmarks, Terasort and Hashjoin.
Spark is an open source cluster computing framework originally developed at UC Berkeley. Intel has made many contributions to Spark's development through code commits, patches, and collaborating with the Spark community. Spark is widely used by companies like Alibaba, Baidu, and Youku for large-scale data analytics and machine learning tasks. It allows for faster iterative jobs than Hadoop through its in-memory computing model and supports multiple workloads including streaming, SQL, and graph processing.
How to use Parquet as a Sasis for ETL and AnalyticsDataWorks Summit
Parquet is a columnar storage format that provides efficient compression and querying capabilities. It aims to store data efficiently for analysis while supporting interoperability across systems. Parquet uses column-oriented storage with efficient encodings and statistics to enable fast querying of large datasets. It integrates with many query engines and frameworks like Hive, Impala, Spark and MapReduce to allow projection and predicate pushdown for optimized queries.
The document discusses saving streaming data from Kafka to S3 using Spark Streaming while ensuring exactly-once delivery. It describes two options for handling failures: (1) writing offsets to a database, requiring additional cleanup; and (2) combining offsets with file paths in S3 to allow overwriting on failure without duplication. The implemented solution uses the second approach by partitioning data by date and sum of starting offsets and deleting folders before writing to ensure exactly-once delivery in a simple way without additional systems.
Spark is a fast and general engine for large-scale data processing. It was designed to be fast, easy to use and supports machine learning. Spark achieves high performance by keeping data in-memory as much as possible using its Resilient Distributed Datasets (RDDs) abstraction. RDDs allow data to be partitioned across nodes and operations are performed in parallel. The Spark architecture uses a master-slave model with a driver program coordinating execution across worker nodes. Transformations operate on RDDs to produce new RDDs while actions trigger job execution and return results.
In this talk, we present a comprehensive framework for assessing the correctness, stability, and performance of the Spark SQL engine. Apache Spark is one of the most actively developed open source projects, with more than 1200 contributors from all over the world. At this scale and pace of development, mistakes bound to happen. To automatically identify correctness issues and performance regressions, we have build a testing pipeline that consists of two complementary stages: randomized testing and benchmarking.
Randomized query testing aims at extending the coverage of the typical unit testing suites, while we use micro and application-like benchmarks to measure new features and make sure existing ones do not regress. We will discuss various approaches we take, including random query generation, random data generation, random fault injection, and longevity stress tests. We will demonstrate the effectiveness of the framework by highlighting several correctness issues we have found through random query generation and critical performance regressions we were able to diagnose within hours due to our automated benchmarking tools.
Similar to Improving SparkSQL Performance by 30%: How We Optimize Parquet Pushdown and Parquet Reader (20)
The document discusses migrating a data warehouse to the Databricks Lakehouse Platform. It outlines why legacy data warehouses are struggling, how the Databricks Platform addresses these issues, and key considerations for modern analytics and data warehousing. The document then provides an overview of the migration methodology, approach, strategies, and key takeaways for moving to a lakehouse on Databricks.
Data Lakehouse Symposium | Day 1 | Part 1Databricks
The world of data architecture began with applications. Next came data warehouses. Then text was organized into a data warehouse.
Then one day the world discovered a whole new kind of data that was being generated by organizations. The world found that machines generated data that could be transformed into valuable insights. This was the origin of what is today called the data lakehouse. The evolution of data architecture continues today.
Come listen to industry experts describe this transformation of ordinary data into a data architecture that is invaluable to business. Simply put, organizations that take data architecture seriously are going to be at the forefront of business tomorrow.
This is an educational event.
Several of the authors of the book Building the Data Lakehouse will be presenting at this symposium.
Data Lakehouse Symposium | Day 1 | Part 2Databricks
The world of data architecture began with applications. Next came data warehouses. Then text was organized into a data warehouse.
Then one day the world discovered a whole new kind of data that was being generated by organizations. The world found that machines generated data that could be transformed into valuable insights. This was the origin of what is today called the data lakehouse. The evolution of data architecture continues today.
Come listen to industry experts describe this transformation of ordinary data into a data architecture that is invaluable to business. Simply put, organizations that take data architecture seriously are going to be at the forefront of business tomorrow.
This is an educational event.
Several of the authors of the book Building the Data Lakehouse will be presenting at this symposium.
The world of data architecture began with applications. Next came data warehouses. Then text was organized into a data warehouse.
Then one day the world discovered a whole new kind of data that was being generated by organizations. The world found that machines generated data that could be transformed into valuable insights. This was the origin of what is today called the data lakehouse. The evolution of data architecture continues today.
Come listen to industry experts describe this transformation of ordinary data into a data architecture that is invaluable to business. Simply put, organizations that take data architecture seriously are going to be at the forefront of business tomorrow.
This is an educational event.
Several of the authors of the book Building the Data Lakehouse will be presenting at this symposium.
The document discusses the challenges of modern data, analytics, and AI workloads. Most enterprises struggle with siloed data systems that make integration and productivity difficult. The future of data lies with a data lakehouse platform that can unify data engineering, analytics, data warehousing, and machine learning workloads on a single open platform. The Databricks Lakehouse platform aims to address these challenges with its open data lake approach and capabilities for data engineering, SQL analytics, governance, and machine learning.
5 Critical Steps to Clean Your Data Swamp When Migrating Off of HadoopDatabricks
In this session, learn how to quickly supplement your on-premises Hadoop environment with a simple, open, and collaborative cloud architecture that enables you to generate greater value with scaled application of analytics and AI on all your data. You will also learn five critical steps for a successful migration to the Databricks Lakehouse Platform along with the resources available to help you begin to re-skill your data teams.
Democratizing Data Quality Through a Centralized PlatformDatabricks
Bad data leads to bad decisions and broken customer experiences. Organizations depend on complete and accurate data to power their business, maintain efficiency, and uphold customer trust. With thousands of datasets and pipelines running, how do we ensure that all data meets quality standards, and that expectations are clear between producers and consumers? Investing in shared, flexible components and practices for monitoring data health is crucial for a complex data organization to rapidly and effectively scale.
At Zillow, we built a centralized platform to meet our data quality needs across stakeholders. The platform is accessible to engineers, scientists, and analysts, and seamlessly integrates with existing data pipelines and data discovery tools. In this presentation, we will provide an overview of our platform’s capabilities, including:
Giving producers and consumers the ability to define and view data quality expectations using a self-service onboarding portal
Performing data quality validations using libraries built to work with spark
Dynamically generating pipelines that can be abstracted away from users
Flagging data that doesn’t meet quality standards at the earliest stage and giving producers the opportunity to resolve issues before use by downstream consumers
Exposing data quality metrics alongside each dataset to provide producers and consumers with a comprehensive picture of health over time
Learn to Use Databricks for Data ScienceDatabricks
Data scientists face numerous challenges throughout the data science workflow that hinder productivity. As organizations continue to become more data-driven, a collaborative environment is more critical than ever — one that provides easier access and visibility into the data, reports and dashboards built against the data, reproducibility, and insights uncovered within the data.. Join us to hear how Databricks’ open and collaborative platform simplifies data science by enabling you to run all types of analytics workloads, from data preparation to exploratory analysis and predictive analytics, at scale — all on one unified platform.
Why APM Is Not the Same As ML MonitoringDatabricks
Application performance monitoring (APM) has become the cornerstone of software engineering allowing engineering teams to quickly identify and remedy production issues. However, as the world moves to intelligent software applications that are built using machine learning, traditional APM quickly becomes insufficient to identify and remedy production issues encountered in these modern software applications.
As a lead software engineer at NewRelic, my team built high-performance monitoring systems including Insights, Mobile, and SixthSense. As I transitioned to building ML Monitoring software, I found the architectural principles and design choices underlying APM to not be a good fit for this brand new world. In fact, blindly following APM designs led us down paths that would have been better left unexplored.
In this talk, I draw upon my (and my team’s) experience building an ML Monitoring system from the ground up and deploying it on customer workloads running large-scale ML training with Spark as well as real-time inference systems. I will highlight how the key principles and architectural choices of APM don’t apply to ML monitoring. You’ll learn why, understand what ML Monitoring can successfully borrow from APM, and hear what is required to build a scalable, robust ML Monitoring architecture.
The Function, the Context, and the Data—Enabling ML Ops at Stitch FixDatabricks
Autonomy and ownership are core to working at Stitch Fix, particularly on the Algorithms team. We enable data scientists to deploy and operate their models independently, with minimal need for handoffs or gatekeeping. By writing a simple function and calling out to an intuitive API, data scientists can harness a suite of platform-provided tooling meant to make ML operations easy. In this talk, we will dive into the abstractions the Data Platform team has built to enable this. We will go over the interface data scientists use to specify a model and what that hooks into, including online deployment, batch execution on Spark, and metrics tracking and visualization.
Stage Level Scheduling Improving Big Data and AI IntegrationDatabricks
In this talk, I will dive into the stage level scheduling feature added to Apache Spark 3.1. Stage level scheduling extends upon Project Hydrogen by improving big data ETL and AI integration and also enables multiple other use cases. It is beneficial any time the user wants to change container resources between stages in a single Apache Spark application, whether those resources are CPU, Memory or GPUs. One of the most popular use cases is enabling end-to-end scalable Deep Learning and AI to efficiently use GPU resources. In this type of use case, users read from a distributed file system, do data manipulation and filtering to get the data into a format that the Deep Learning algorithm needs for training or inference and then sends the data into a Deep Learning algorithm. Using stage level scheduling combined with accelerator aware scheduling enables users to seamlessly go from ETL to Deep Learning running on the GPU by adjusting the container requirements for different stages in Spark within the same application. This makes writing these applications easier and can help with hardware utilization and costs.
There are other ETL use cases where users want to change CPU and memory resources between stages, for instance there is data skew or perhaps the data size is much larger in certain stages of the application. In this talk, I will go over the feature details, cluster requirements, the API and use cases. I will demo how the stage level scheduling API can be used by Horovod to seamlessly go from data preparation to training using the Tensorflow Keras API using GPUs.
The talk will also touch on other new Apache Spark 3.1 functionality, such as pluggable caching, which can be used to enable faster dataframe access when operating from GPUs.
Simplify Data Conversion from Spark to TensorFlow and PyTorchDatabricks
In this talk, I would like to introduce an open-source tool built by our team that simplifies the data conversion from Apache Spark to deep learning frameworks.
Imagine you have a large dataset, say 20 GBs, and you want to use it to train a TensorFlow model. Before feeding the data to the model, you need to clean and preprocess your data using Spark. Now you have your dataset in a Spark DataFrame. When it comes to the training part, you may have the problem: How can I convert my Spark DataFrame to some format recognized by my TensorFlow model?
The existing data conversion process can be tedious. For example, to convert an Apache Spark DataFrame to a TensorFlow Dataset file format, you need to either save the Apache Spark DataFrame on a distributed filesystem in parquet format and load the converted data with third-party tools such as Petastorm, or save it directly in TFRecord files with spark-tensorflow-connector and load it back using TFRecordDataset. Both approaches take more than 20 lines of code to manage the intermediate data files, rely on different parsing syntax, and require extra attention for handling vector columns in the Spark DataFrames. In short, all these engineering frictions greatly reduced the data scientists’ productivity.
The Databricks Machine Learning team contributed a new Spark Dataset Converter API to Petastorm to simplify these tedious data conversion process steps. With the new API, it takes a few lines of code to convert a Spark DataFrame to a TensorFlow Dataset or a PyTorch DataLoader with default parameters.
In the talk, I will use an example to show how to use the Spark Dataset Converter to train a Tensorflow model and how simple it is to go from single-node training to distributed training on Databricks.
Scaling your Data Pipelines with Apache Spark on KubernetesDatabricks
There is no doubt Kubernetes has emerged as the next generation of cloud native infrastructure to support a wide variety of distributed workloads. Apache Spark has evolved to run both Machine Learning and large scale analytics workloads. There is growing interest in running Apache Spark natively on Kubernetes. By combining the flexibility of Kubernetes and scalable data processing with Apache Spark, you can run any data and machine pipelines on this infrastructure while effectively utilizing resources at disposal.
In this talk, Rajesh Thallam and Sougata Biswas will share how to effectively run your Apache Spark applications on Google Kubernetes Engine (GKE) and Google Cloud Dataproc, orchestrate the data and machine learning pipelines with managed Apache Airflow on GKE (Google Cloud Composer). Following topics will be covered: – Understanding key traits of Apache Spark on Kubernetes- Things to know when running Apache Spark on Kubernetes such as autoscaling- Demonstrate running analytics pipelines on Apache Spark orchestrated with Apache Airflow on Kubernetes cluster.
Scaling and Unifying SciKit Learn and Apache Spark PipelinesDatabricks
Pipelines have become ubiquitous, as the need for stringing multiple functions to compose applications has gained adoption and popularity. Common pipeline abstractions such as “fit” and “transform” are even shared across divergent platforms such as Python Scikit-Learn and Apache Spark.
Scaling pipelines at the level of simple functions is desirable for many AI applications, however is not directly supported by Ray’s parallelism primitives. In this talk, Raghu will describe a pipeline abstraction that takes advantage of Ray’s compute model to efficiently scale arbitrarily complex pipeline workflows. He will demonstrate how this abstraction cleanly unifies pipeline workflows across multiple platforms such as Scikit-Learn and Spark, and achieves nearly optimal scale-out parallelism on pipelined computations.
Attendees will learn how pipelined workflows can be mapped to Ray’s compute model and how they can both unify and accelerate their pipelines with Ray.
Sawtooth Windows for Feature AggregationsDatabricks
In this talk about zipline, we will introduce a new type of windowing construct called a sawtooth window. We will describe various properties about sawtooth windows that we utilize to achieve online-offline consistency, while still maintaining high-throughput, low-read latency and tunable write latency for serving machine learning features.We will also talk about a simple deployment strategy for correcting feature drift – due operations that are not “abelian groups”, that operate over change data.
We want to present multiple anti patterns utilizing Redis in unconventional ways to get the maximum out of Apache Spark.All examples presented are tried and tested in production at Scale at Adobe. The most common integration is spark-redis which interfaces with Redis as a Dataframe backing Store or as an upstream for Structured Streaming. We deviate from the common use cases to explore where Redis can plug gaps while scaling out high throughput applications in Spark.
Niche 1 : Long Running Spark Batch Job – Dispatch New Jobs by polling a Redis Queue
· Why?
o Custom queries on top a table; We load the data once and query N times
· Why not Structured Streaming
· Working Solution using Redis
Niche 2 : Distributed Counters
· Problems with Spark Accumulators
· Utilize Redis Hashes as distributed counters
· Precautions for retries and speculative execution
· Pipelining to improve performance
Re-imagine Data Monitoring with whylogs and SparkDatabricks
In the era of microservices, decentralized ML architectures and complex data pipelines, data quality has become a bigger challenge than ever. When data is involved in complex business processes and decisions, bad data can, and will, affect the bottom line. As a result, ensuring data quality across the entire ML pipeline is both costly, and cumbersome while data monitoring is often fragmented and performed ad hoc. To address these challenges, we built whylogs, an open source standard for data logging. It is a lightweight data profiling library that enables end-to-end data profiling across the entire software stack. The library implements a language and platform agnostic approach to data quality and data monitoring. It can work with different modes of data operations, including streaming, batch and IoT data.
In this talk, we will provide an overview of the whylogs architecture, including its lightweight statistical data collection approach and various integrations. We will demonstrate how the whylogs integration with Apache Spark achieves large scale data profiling, and we will show how users can apply this integration into existing data and ML pipelines.
Raven: End-to-end Optimization of ML Prediction QueriesDatabricks
Machine learning (ML) models are typically part of prediction queries that consist of a data processing part (e.g., for joining, filtering, cleaning, featurization) and an ML part invoking one or more trained models. In this presentation, we identify significant and unexplored opportunities for optimization. To the best of our knowledge, this is the first effort to look at prediction queries holistically, optimizing across both the ML and SQL components.
We will present Raven, an end-to-end optimizer for prediction queries. Raven relies on a unified intermediate representation that captures both data processing and ML operators in a single graph structure.
This allows us to introduce optimization rules that
(i) reduce unnecessary computations by passing information between the data processing and ML operators
(ii) leverage operator transformations (e.g., turning a decision tree to a SQL expression or an equivalent neural network) to map operators to the right execution engine, and
(iii) integrate compiler techniques to take advantage of the most efficient hardware backend (e.g., CPU, GPU) for each operator.
We have implemented Raven as an extension to Spark’s Catalyst optimizer to enable the optimization of SparkSQL prediction queries. Our implementation also allows the optimization of prediction queries in SQL Server. As we will show, Raven is capable of improving prediction query performance on Apache Spark and SQL Server by up to 13.1x and 330x, respectively. For complex models, where GPU acceleration is beneficial, Raven provides up to 8x speedup compared to state-of-the-art systems. As part of the presentation, we will also give a demo showcasing Raven in action.
Processing Large Datasets for ADAS Applications using Apache SparkDatabricks
Semantic segmentation is the classification of every pixel in an image/video. The segmentation partitions a digital image into multiple objects to simplify/change the representation of the image into something that is more meaningful and easier to analyze [1][2]. The technique has a wide variety of applications ranging from perception in autonomous driving scenarios to cancer cell segmentation for medical diagnosis.
Exponential growth in the datasets that require such segmentation is driven by improvements in the accuracy and quality of the sensors generating the data extending to 3D point cloud data. This growth is further compounded by exponential advances in cloud technologies enabling the storage and compute available for such applications. The need for semantically segmented datasets is a key requirement to improve the accuracy of inference engines that are built upon them.
Streamlining the accuracy and efficiency of these systems directly affects the value of the business outcome for organizations that are developing such functionalities as a part of their AI strategy.
This presentation details workflows for labeling, preprocessing, modeling, and evaluating performance/accuracy. Scientists and engineers leverage domain-specific features/tools that support the entire workflow from labeling the ground truth, handling data from a wide variety of sources/formats, developing models and finally deploying these models. Users can scale their deployments optimally on GPU-based cloud infrastructure to build accelerated training and inference pipelines while working with big datasets. These environments are optimized for engineers to develop such functionality with ease and then scale against large datasets with Spark-based clusters on the cloud.
Massive Data Processing in Adobe Using Delta LakeDatabricks
At Adobe Experience Platform, we ingest TBs of data every day and manage PBs of data for our customers as part of the Unified Profile Offering. At the heart of this is a bunch of complex ingestion of a mix of normalized and denormalized data with various linkage scenarios power by a central Identity Linking Graph. This helps power various marketing scenarios that are activated in multiple platforms and channels like email, advertisements etc. We will go over how we built a cost effective and scalable data pipeline using Apache Spark and Delta Lake and share our experiences.
What are we storing?
Multi Source – Multi Channel Problem
Data Representation and Nested Schema Evolution
Performance Trade Offs with Various formats
Go over anti-patterns used
(String FTW)
Data Manipulation using UDFs
Writer Worries and How to Wipe them Away
Staging Tables FTW
Datalake Replication Lag Tracking
Performance Time!
Harnessing Wild and Untamed (Publicly Available) Data for the Cost efficient ...weiwchu
We recently discovered that models trained with large-scale speech datasets sourced from the web could achieve superior accuracy and potentially lower cost than traditionally human-labeled or simulated speech datasets. We developed a customizable AI-driven data labeling system. It infers word-level transcriptions with confidence scores, enabling supervised ASR training. It also robustly generates phone-level timestamps even in the presence of transcription or recognition errors, facilitating the training of TTS models. Moreover, It automatically assigns labels such as scenario, accent, language, and topic tags to the data, enabling the selection of task-specific data for training a model tailored to that particular task. We assessed the effectiveness of the datasets by fine-tuning open-source large speech models such as Whisper and SeamlessM4T and analyzing the resulting metrics. In addition to openly-available data, our data handling system can also be tailored to provide reliable labels for proprietary data from certain vertical domains. This customization enables supervised training of domain-specific models without the need for human labelers, eliminating data breach risks and significantly reducing data labeling cost.
Towards an Analysis-Ready, Cloud-Optimised service for FAIR fusion dataSamuel Jackson
We present our work to improve data accessibility and performance for data-intensive tasks within the fusion research community. Our primary goal is to develop services that facilitate efficient access for data-intensive applications while ensuring compliance with FAIR principles [1], as well as adoption of interoperable tools, methods and standards.
The major outcome of our work is the successful creation and deployment of a data service for the MAST (Mega Ampere Spherical Tokamak) experiment [2], leading to substantial enhancements in data discoverability, accessibility, and overall data retrieval performance, particularly in scenarios involving large-scale data access. Our work follows the principles of Analysis-Ready, Cloud Optimised (ARCO) data [3] by using cloud optimised data formats for fusion data.
Our system consists of a query-able metadata catalogue, complemented with an object storage system for publicly serving data from the MAST experiment. We will show how our solution integrates with the Pandata stack [4] to enable data analysis and processing at scales that would have previously been intractable, paving the way for data-intensive workflows running routinely with minimal pre-processing on the part of the researcher. By using a cloud-optimised file format such as zarr [5] we can enable interactive data analysis and visualisation while avoiding large data transfers. Our solution integrates with common python data analysis libraries for large, complex scientific data such as xarray [6] for complex data structures and dask [7] for parallel computation and lazily working with larger that memory datasets.
The incorporation of these technologies is vital for advancing simulation, design, and enabling emerging technologies like machine learning and foundation models, all of which rely on efficient access to extensive repositories of high-quality data. Relying on the FAIR guiding principles for data stewardship not only enhances data findability, accessibility, and reusability, but also fosters international cooperation on the interoperability of data and tools, driving fusion research into new realms and ensuring its relevance in an era characterised by advanced technologies in data science.
[1] Wilkinson, M., Dumontier, M., Aalbersberg, I. et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci Data 3, 160018 (2016) https://doi.org/10.1038/sdata.2016.18
[2] M Cox, The Mega Amp Spherical Tokamak, Fusion Engineering and Design, Volume 46, Issues 2–4, 1999, Pages 397-404, ISSN 0920-3796, https://doi.org/10.1016/S0920-3796(99)00031-9
[3] Stern, Charles, et al. "Pangeo forge: crowdsourcing analysis-ready, cloud optimized data production." Frontiers in Climate 3 (2022): 782909.
[4] Bednar, James A., and Martin Durant. "The Pandata Scalable Open-Source Analysis Stack." (2023).
[5] Alistair Miles (2024) ‘zarr-developers/zarr-python: v2.17.1’. Zenodo. doi: 10.5281/zenodo.10790679
[6] Hoyer, S. & Hamman, J., (20
Annex K RBF's The World Game pdf documentSteven McGee
Signals & Telemetry Annex K for RBF's The World Game / Trade Federations / USPTO 13/573,002 Heart Beacon Cycle Time - Space Time Chain meters, metrics, standards. Adaptive Procedural template framework structured data derived from DoD / NATO's system of systems engineering tech framework
Combined supervised and unsupervised neural networks for pulse shape discrimi...Samuel Jackson
Our methodology for pulse shape discrimination is split into two steps. Firstly, we learn a model to discriminate between pulses using "clean" low-rate examples by removing pile-up & saturated events. In addition to traditional tail sum discrimination, we investigate three different choices for discrimination between γ-pulses, fast, thermal neutrons. We consider clustering the pulses directly using Gaussian Mixture Modelling (GMM), using variational autoencoders to learn a representation of the pulses and then clustering the learned representation (VAE+GMM) and using density ratio estimation to discriminate between a mixed (γ + neutron) and pure (γ only) sources using a multi-layer perceptron (MLP) as a supervised learning problem.
Secondly, we aim to classify and recover pile-up events in the < 150 ns regime by training a single unified multi-label MLP. To frame the problem as a multi-label supervised learning method, we first simulate pile-up events with known components. Then, using the simulated data and combining it with single event data, we train a final multi-label MLP to output a binary code indicating both how many and which type of events are present within an event window.
Overview of Statistical software such as ODK, surveyCTO,and CSPro
2. Software installation(for computer, and tablet or mobile devices)
3. Create a data entry application
4. Create the data dictionary
5. Create the data entry forms
6. Enter data
7. Add Edits to the Data Entry Application
8. CAPI questions and texts
Introduction to Data Science
1.1 What is Data Science, importance of data science,
1.2 Big data and data Science, the current Scenario,
1.3 Industry Perspective Types of Data: Structured vs. Unstructured Data,
1.4 Quantitative vs. Categorical Data,
1.5 Big Data vs. Little Data, Data science process
1.6 Role of Data Scientist
Improving SparkSQL Performance by 30%: How We Optimize Parquet Pushdown and Parquet Reader
1. Improving Spark SQL Performance by 30%:
How We Optimize Parquet Filter Pushdown
and Parquet Reader
Ke Sun (sunke3296@gmail.com)
Senior Engineer of Data Engine Team, ByteDance
2. Who We Are
l Data Engine team of ByteDance
l Build a platform of one-stop
experience for OLAP , on which
users can analyze EB level data by
writing SQL without caring about
the underlying execution engine
3. What We Do
l Manage Spark SQL / Presto / Hive
workload
l Offer open API and serverless OLAP
platform
l Optimize Spark SQL / Presto / Hudi /
Hive engine
l Design data architecture for most
business lines in ByteDance
4. Agenda
Spark SQL at ByteDance
How Spark Read Parquet
Optimization of Parquet Filter
Pushdown and Parquet Reader
at ByteDance
6. 2017 2019 20202016 2018
Spark SQL at ByteDance
Small Scale Experiments
Ad-hoc Workload
Few ETL Workload
Full-production deployment & migration
Main engine in DW area
7. Spark SQL at ByteDance
▪ Spark SQL covers 98%+ ETL workload
▪ Parquet is the default file format in data warehouse and
vectorizedReader is also enabled by default
9. How Spark Read Parquet
▪ Overview of Parquet
▪ Procedure of Parquet Reading
▪ What We can Optimize
10. How Spark Read Parquet
▪ Overview of Parquet
Column Pruning
More efficient compression
Parquet can skip useless data by spark filter
pushdown using Footer & RowGroup statistics
https://raw.github.com/apache/parquet-format/master/doc/images/FileLayout.gif
11. How Spark Read Parquet
▪ Procedure of Parquet Reading
▪ VectorizedParquetRecordReader skips useless
RowGroups by filter pushdown translated by
ParquetFilters
▪ VectorizedParquetRecordReader builds column readers
for every target column and these column readers read
data together in batch
DataSourceScanExec
.inputRDD
ParquetFileFormat
.buildReaderWithPartitionValues
VectorizedParquetRecordReader
. nextBatch
12. How Spark Read Parquet
▪ Optimization – Statistics are not distinguishable
select * from table_name where date = ‘***’ and category = ‘test’
(date is partition column and category is predicate column)
For example, Spark reads all the 3 RowGroups for statistics are not distinguishable.
min of category max of category read or not
RowGroup1 a1 z1 Yes
RowGroup2 a2 z2 Yes
RowGroup3 a3 z3 Yes
Min/Max Statistics of RowGroup for a Parquet File
13. How Spark Read Parquet
▪ Optimization – Statistics are not distinguishable
Parquet filter pushdown works poorly when predicate columns are out of order in
parquet files and this phenomenon is reasonable
It is valuable to sort the common used predicate columns in parquet files to
reduce IO
14. How Spark Read Parquet
▪ Optimization – Spark reads too much unnecessary data
select col1 from table_name where date = ‘***’ and col2 = ‘test’
col1 col2 col3
col1 col2 col3
col1 col2 col3
ParquetFile
RowGroup1
RowGroup2
RowGroup3
▪ RowGroup1 is skipped by filter pushdown
▪ Col3 is skipped by column pruning
▪ Col1 and col2 are read together by vectorized reader
15. How Spark Read Parquet
▪ Optimization – Spark reads too much unnecessary data
select col1 from table_name where date = ‘***’ and col2 = ‘test’
col1 col2 col3
col1 col2 col3
col1 col2 col3
ParquetFile
RowGroup1
RowGroup2
RowGroup3
▪ Most data of col1 is unnecessary to read
if filter ratio (col2=‘test’) is very high
▪ It is valuable to read & filter data by filter columns firstly
and then read data of other columns
17. Optimization of Parquet Filter Pushdown
▪ Statistics are not distinguishable
▪ Increase parquet statistics discrimination
▪ Low Overhead: sort all data is expensive or even impossible
▪ Automation: users do not need to update ETL jobs
LocalSort: Add a SortExec node before InsertIntoHiveTable node
18. ▪ Which columns to be sorted?
▪ Analyze the history queries and choose the most common used predicate columns
▪ Configure sort columns to table property of hive table and Spark SQL will read this property
▪ It is a automatic procedure without manual intervention
Optimization of Parquet Filter Pushdown
…
Project
SortExec
…
Project
InsertIntoHiveTable
InsertIntoHiveTable
19. Optimization of Parquet Filter Pushdown
▪ Spark reads less data for more efficient statistics
▪ Parquet file size is much smaller for data is more efficient to compress
▪ Only near 5% overhead
Spark reads only one RowGroup after sorting data by column category
min of category max of category read or not
RowGroup1 a1 g1 No
RowGroup2 g2 u2 Yes
RowGroup3 u3 z3 No
Min/Max Statistics of RowGroup for a Parquet File
20. Optimization of Parquet Reader
▪ Spark reads too much unnecessary data
▪ Filter unnecessary data as soon as possible
Prewhere: Read data of filter columns in batch firstly and skip other
columns if unmatched (It is a good idea from ClickHouse)
21. Optimization of Parquet Reader
▪ Split parquet reader into 2 reader: FilterReader for filter
columns and NonFilterReader for other columns
col1 col2 col3
col1 col2 col3
col1 col2 col3
RowGroup1
RowGroup2
RowGroup3
VectorizedParquetRecordReader
col1 col2 col3
col1 col2 col3
col1 col2 col3
FilterReader
col1 col2 col3
col1 col2 col3
col1 col2 col3
NonFilterReader
PrewhereVectorizedParquetRecordReader
22. Optimization of Parquet Reader
▪ Skip Page
▪ Skip Decoding
▪ Skip RowGroup
Potential BenefitFilterReader reads data in batch
Apply filter expressions to data
match
NonFilterReader reads data in
batch and skip unnecessary data
N
Y
Union data and return in batch
23. Optimization of Parquet Reader
▪ ByteType
▪ ShortType
▪ IntegerType
▪ LongType
▪ FloatType
▪ DoubleType
▪ StringType
▪ >
▪ >=
▪ <
▪ <=
▪ =
▪ In
▪ isNull
▪ isNotNull
Supported Filter TypeSupported Data Type of Filter Column