A Developer’s View into Spark's Memory Model with Wenchen Fan
•Als PPTX, PDF herunterladen•
8 gefällt mir•1,541 views
As part of Project Tungsten, we started an ongoing effort to substantially improve the memory and CPU efficiency of Apache Spark’s backend execution and push performance closer to the limits of modern hardware. In this talk, we’ll take a deep dive into Apache Spark’s unified memory model and discuss how Spark exploits memory hierarchy and leverages application semantics to manage memory explicitly (both on and off-heap) to eliminate the overheads of JVM object model and garbage collection.
Empfohlen
Empfohlen
Weitere ähnliche Inhalte
Was ist angesagt?
Was ist angesagt? (20)
Ähnlich wie A Developer’s View into Spark's Memory Model with Wenchen Fan
Ähnlich wie A Developer’s View into Spark's Memory Model with Wenchen Fan (20)
Mehr von Databricks
Mehr von Databricks (20)
Kürzlich hochgeladen
Kürzlich hochgeladen (20)
A Developer’s View into Spark's Memory Model with Wenchen Fan
- 1. A Developer’s View Into Spark’s Memory Model Wenchen Fan 2017-6-7
- 2. About Me • Software Engineer @ • Apache Spark Committer • One of the most active Spark contributors
- 4. About Databricks TEAM Started Spark project (now Apache Spark) at UC Berkeley in 2009 MISSON Make Big Data Simple PRODUC TUnified Analytics Platform
- 6. Memory Model inside Executor data 1010100001010 0010001001001 1010001000111 1010100100101 internal format Cache Manager Memory Manager operators allocate memor y memor y pages allocate memor ymemor y pages
- 7. Memory Model inside Executor data 1010100001010 0010001001001 1010001000111 1010100100101 Cache Manager Memory Manager operators allocate memor y memor y pages allocate memor ymemor y pages internal format
- 8. Memory Allocation • Allocation happens in page granularity. • Off-heap supported! • Page is not fixed-size, but has a lower and upper bound. • No pooling, pages are freed once there is no data on it.
- 9. Why var-length page and no pooling? • Pros: • simplify the implementation. (no single record will across pages) • free memory immediately so that the OS can use them for file buffer, etc. • Cons: • can not handle super big single record. (very rare in reality) • fragmentation for records bigger than page size lower bound. (the lower bound is several mega bytes, so it’s also rare) • overhead in allocation. (most malloc algorithms should work well)
- 10. Memory Model inside Executor data 1010100001010 0010001001001 1010001000111 1010100100101 Cache Manager Memory Manager operators allocate memor y memor y pages allocate memor ymemor y pages internal format
- 11. Java Objects Based Row Format Array BoxedInteger(123) String(“data”) String(“bricks”) Row • 5+ objects • high space overhead • slow value accessing • expensive hashCode() (123, “data”, “bricks”)
- 12. Data objects? No! • It is hard to monitor and control the memory usage when we have a lot of objects. • Garbage collection will be the killer. • High serialization cost when transfer data inside cluster.
- 13. Efficient Binary Format 0x0 null tracking (123, “data”, “bricks”)
- 14. Efficient Binary Format 0x0 123 null tracking (123, “data”, “bricks”)
- 15. Efficient Binary Format 0x0 123 32 “data”4 null tracking (123, “data”, “bricks”) offset and length of
- 16. Efficient Binary Format “bricks”0x0 123 32 40 “data”4 6 null tracking (123, “data”, “bricks”) offset and length of offset and length of data
- 17. Memory Model inside Executor data 1010100001010 0010001001001 1010001000111 1010100100101 Cache Manager Memory Manager operators allocate memor y memor y pages allocate memor ymemor y pages internal format
- 18. Operate On Binary 0x 0 12 3 2 4 4 “data ” 0x 0 12 3 2 4 4 “data ” 0x 0 1 6 2 “da” 123 > JSON files substring
- 19. How to process binary data more efficiently?
- 20. Understanding CPU Cache Memory is becoming slower and slower than CPU.
- 21. Understanding CPU Cache Pre-fetch frequently accessed data into CPU cache.
- 22. The most 2 important algorithms in big data are ... Sort and Hash!
- 27. Naive Sort Each comparison needs to access 2 different memory regions, which makes it hard for CPU cache to pre-fetch data, poor cache locality!
- 31. Cache-aware Sort Most of the time, just go through the key-prefixes in a linear fashion, good cache locality!
- 32. Naive Hash Map key1pointer v1 key2pointer v2 key3pointer v3
- 33. Naive Hash Map pointer pointer pointer key3 lookup key1 v1 key2 v2 key3 v3
- 34. Naive Hash Map pointer pointer pointer key3 hash(key) % size key1 v1 key2 v2 key3 v3
- 35. Naive Hash Map pointer pointer pointer key3 compare these 2 keys key1 v1 key2 v2 key3 v3
- 36. Naive Hash Map pointer pointer pointer key3 quadratic probing key1 v1 key2 v2 key3 v3
- 37. Naive Hash Map key3 compare these 2 keys pointer pointer pointer key1 v1 key2 v2 key3 v3
- 38. Naive Hash Map Each lookup needs many pointer dereferences and key comparison when hash collision happens, and jumps between 2 memory regions, bad cache locality!
- 39. ptr ptr ptr full hash full hash full hash Cache-aware Hash Map key1 v1 key2 v2 key3 v3
- 40. ptr ptr ptr full hash full hash full hash Cache-aware Hash Map key3 lookup key1 v1 key2 v2 key3 v3
- 41. ptr ptr ptr full hash full hash full hash Cache-aware Hash Map key3 hash(key) % size, and compare the full hash key1 v1 key2 v2 key3 v3
- 42. ptr ptr ptr full hash full hash full hash Cache-aware Hash Map key3 quadratic probing, and compare the full hash key1 v1 key2 v2 key3 v3
- 43. ptr ptr ptr full hash full hash full hash Cache-aware Hash Map key3 key1 v1 key2 v2 key3 v3 compare these 2 keys
- 44. Cache-aware Hash Map Each lookup mostly only needs one pointer dereference and key comparison(full hash collision is rare), and access data mostly in a single memory region, better cache locality!
- 45. Recap: Cache-aware data structure How to improve cache locality … • store key-prefix with pointer. • store key full hash with pointer. Store extra information to try to keep the memory accessing in a single region.
- 46. Memory Model inside Executor data 1010100001010 0010001001001 1010001000111 1010100100101 Cache Manager Memory Manager operators allocate memor y memor y pages allocate memor ymemor y pages internal format
- 47. Future Work • SPARK-19489: Stable serialization format for external & native code integration. • SPARK-15689: Data source API v2 • SPARK-15687: Columnar execution engine.
- 48. Try Apache Spark in Databricks! UNIFIED ANALYTICS PLATFORM • Collaborative cloud environment • Free version (community edition) DATABRICKS RUNTIME 3.0 • Apache Spark - optimized for the cloud • Caching and optimization layer - DBIO • Enterprise security - DBES Try for free today. databricks.com
- 49. Thank You
Hinweis der Redaktion
- I’m not lying!
- first of all, spark is a distributed engine, let’s see the memory usage at the cluster level. a typical spark cluster has one master node and several worker nodes the entries of spark application is the driver process, it can be run on any node. when driver launches, it will apply for executors from master master will ask workers to launch executors for a driver each executor is a JVM process and has a memory manager and a thread pool to run tasks distribute resource across spark applications is out of this talk this talk focus on memory management inside executors Let’s zoom in!
- Keep data as binary and operate on binary. execution and storage are the 2 major memory consumers in Spark. Memory manager only manages part of the total available memory.
- 16g upper bound
- why use internal format? why not just use objects?
- some more reasons about why Spark doesn’t use data objects
- how to operate on binary data directly.
- The whole data flow looks efficient, but, in Spark, we keep asking ourself, can it run faster?
- shall we re-implement all operators in Spark to be CPU cache friendly? That’s a lot of work! Instead, we only focus on 2 algorithms, because …
- many fancy algorithms/systems are fundamentally based on sort and hash, improving these 2 algorithms can benefit a lot of operators in Spark.
- most sort algorithms work by, picking 2 records, compare and switch, and repeat that again and again
- then repeat
- The idea behind this algorithm is that, mostly we can compare 2 records with first several bytes.
- collision is common
- For a hash table, collision happens frequently when you have many entries.
- the idea behind this algorithm is that, although collision is common in hash table, but full hash value collision is very rare. mostly we can identify a key with its full hash value.
- The data flow is efficient, but the beginning of it, the data source may not. data source is a public API, RDD[Row]