Download as PDF, PPTX
![USING DRIZZLE API
DAGScheduler.scala
def runJob(
rdd: RDD[T],
func: Iterator[T] => U)
Internal
def runBatchJob(
rdds: Seq[RDD[T]],
funcs: Seq[Iterator[T] => U])](https://image.slidesharecdn.com/1ppandavendataraman-160614191233/85/Low-Latency-Execution-For-Apache-Spark-28-320.jpg)
![USING DRIZZLE API
DAGScheduler.scala
def runJob(
rdd: RDD[T],
func: Iterator[T] => U)
Internal
def runBatchJob(
rdds: Seq[RDD[T]],
funcs: Seq[Iterator[T] => U])](https://image.slidesharecdn.com/1ppandavendataraman-160614191233/75/Low-Latency-Execution-For-Apache-Spark-28-2048.jpg)
More Related Content
Similar to Low Latency Execution For Apache Spark
![RTP_AR_Basic_Learners' Workbook_KS2 [FOR REPRODUCTION] (1).pdf](https://cdn.slidesharecdn.com/ss_thumbnails/rtparbasiclearnersworkbookks2forreproduction1-251016024943-e51a16ac-thumbnail.jpg?width=600ounds&width=560&fit=bounds)
![Agentic Systems and Compliance - A brief intro [1.2]](https://cdn.slidesharecdn.com/ss_thumbnails/agenticsystemsandcompliace-1-251018025303-958a42ec-thumbnail.jpg?width=600ounds&width=560&fit=bounds)
Low Latency Execution For Apache Spark
- 1. Low latency executionfor apache spark Shivaram Venkataraman, Aurojit Panda, Kay Ousterhout
- 2. Who am I? PhD candidate, AMPLab UC Berkeley Dissertation: System design for large scale machine learning Apache Spark PMC Member. Contributions to Spark core, MLlib, SparkR
- 3. Low latency: SPARKSTREAMING
- 4. Low latency: SPARKSTREAMING
- 5. Low latency: SPARKSTREAMING
- 6. Low latency: execution Scheduler Fewmilliseconds Large Clusters
- 7. Low latency executionengine for iterative workloads Machine learning Stream processing State This talk
- 8.
- 9. Centralized task scheduling Lineage, Parallel Recovery MicrosoftDryad Execution models: batch processing Task Control Message Driver Driver: Coordinate shuffles, data sharing S H U F F L E Network Transfer Iteration B R O A D C A S T
- 10. Execution models: paralleloperators Long-lived operators Checkpointing based fault tolerance Naiad S H U F F L E S H U F F L E Iteration Task Control Message Driver Network Transfer Low Latency
- 11. SCALING behavior 0 50 100 150 200 250 4 8 16 32 64 128 Time(ms) Machines Network Wait Compute Task Fetch Scheduler Delay Cluster: 4 core, r3.xlarge machines Workload: Sum of 10k numbers per-core Median-task time breakdown
- 12. Can we achievelow latency with Apache Spark ?
- 13. DESIGN INSIGHT Fine-grained executionwith coarse-grained scheduling
- 14. DRIZZLE S H U F F L E B R O A D C A S T Iteration Batch Scheduling Pre-SchedulingShuffles Distributed Shared Variables
- 15. DAG scheduling Assign tasksto hosts using (a) locality preferences (b) straggler mitigation (c) fair sharing etc. Tasks Host1 Host2 Driver Host1 Host2 Serialize & Launch Host Metadata Scheduler Same DAG structure for many iterations Can reuse scheduling decisions
- 16. Batch scheduling Schedule abatch of iterations at once Fault tolerance, scheduling at batch boundaries 1 stage in each iteration b = 2
- 17. How much doesthis help ? 1 10 100 1000 4 8 16 32 64 128 Time/Iter(ms) Machines Apache Spark Drizzle-10 Drizzle-50 Drizzle-100 Workload: Sum of 10k numbers per-core Single Stage Job, 100 iterations – Varying Drizzle batch size
- 18. DRIZZLE S H U F F L E B R O A D C A S T Iteration Batch Scheduling Pre-SchedulingShuffles Distributed Shared Variables
- 19. coordinating shuffles: APACHESPARK Task Control Message Data Message Driver Intermediate Data Driver sends metadata Tasks pull data
- 20. coordinating shuffles: PRE-SCHEDULING Pre-scheduledown-stream tasks on executors Trigger tasks once dependencies are met Task Control Message Data Message Driver Intermediate Data Pre-scheduled task
- 21. Push-metadata, pull-data Coalesce metadataacross cores Transfer during downstream stage Data Message Intermediate Data Metadata Message
- 22. Micro-benchmark: 2-stages 0 50 100 150 200 250 300 350 4 8 16 32 64 128 Time/Iter(ms) Machines Apache Spark Drizzle-Pull Workload: Sum of 10k numbers per-core 100 iterations
- 23. DRIZZLE S H U F F L E B R O A D C A S T Iteration Batch Scheduling Pre-SchedulingShuffles Distributed Shared Variables
- 24. Shared variables Fine-grained updatesto shared data Distributed shared variables - MPI using MPI_AllReduce - Bloom, CRDTs - Parameter servers, key-value stores - reduce followed by broadcast
- 25. Drizzle: Replicated sharedvariables Custom merge strategies 1 Commutative updates e.g. vector addition Task Shared Variable M E R G E 1 1 1 1 2 2 2 2
- 26. Enabling async updates Synchronous Asynchronoussemantics within a batch Synchronous semantics enforced at batch boundaries Task Shared Variable 1 2 Async Stale versions 1 1 2 3 Merge
- 27. Evaluation Micro-benchmarks - Single stage - Multiple stages - Shared variables End-to-end experiments - Streaming benchmarks - Logistic Regression Implemented on Apache Spark 1.6.1 Integrations with Spark Streaming, MLlib
- 28. USING DRIZZLE API DAGScheduler.scala def runJob( rdd: RDD[T], func: Iterator[T] => U) Internal def runBatchJob( rdds: Seq[RDD[T]], funcs: Seq[Iterator[T] => U])
- 29. Using drizzle StreamingContext.scala def this( sc: SparkContext, batchDuration: Duration) Spark Streaming def this( sc: SparkContext, batchDuration: Duration, jobsPerBatch: Int)
- 30. b=1 à Batchprocessing Choosing batch size Higher overhead Smaller window for fault tolerance In practice: Largest batch such that overhead is below fixed threshold e.g., For 128 machines, batch of few seconds is enough b=N à Parallel operators Lower overhead Larger window for fault tolerance
- 31. Evaluation Micro-benchmarks - Single stage - Multiple stages - Shared variables End-to-end experiments - Streaming benchmarks - Logistic Regression Implemented on Apache Spark 1.6.1 Integrations with Spark Streaming, MLlib
- 32. Streaming BENCHMARK 0 25 50 75 100 0 300 600 900 1200 1500 CDF Event Latency (ms) Drizzle, 90ms Spark Streaming 560ms Yahoo Streaming Benchmark: 1M JSON Ad-events / second, 64 machines Event Latency: Difference between window end, processing end
- 33. Mllib: LOGISTIC REGRESSION 0 100 200 300 400 500 600 4 8 16 32 64 128 Time/Iter(ms) Machines Apache Spark Drizzle Sparse Logistic Regression on RCV1 677K examples, 47K features
- 34. Work In progress Automaticbatch size tuning Open source release Apache Spark JIRA to discuss potential contribution
- 35. conclusion Overheads when usingApache Spark for streaming, ML workloads Drizzle: Low Latency Execution Engine - Decouple execution from centralized scheduling - Milliseconds latency for iterative workloads Workloads / Questions / Contributions Shivaram Venkataraman shivaram@cs.berkeley.edu