SAP HANA - Main Memory Technology: A Challenge for Development of Business Applications. Jürgen Primsch, SAP AG July 2011

Transcription

1 SAP HANA - Main Memory Technology: A Challenge for Development of Business Applications Jürgen Primsch, SAP AG July 2011

2 Why In-Memory? Information at the Speed of Thought Imagine access to business data, supporting analytical queries on the transactional data, no need for data replication, and zero response time How would this change the way you work with data? How would this change your perception of information availability? How would this new information influence your decisions? In-memory databases are a technology with huge disruption potential, we need to assume a competitive leadership position SAP AG. All rights reserved. 2

3 Motivation: Make Use of Modern Hardware Game Changing Hardware Trends Disk is Tape Main memory affordable CPU: more cores, no clock rate increase Software makers need to react Past CPU clock rate growing Software runs faster without change Today CPU clock rate growth is flat Number of cores increases Write software that scales with number of cores! 2011 SAP AG. All rights reserved. 3

4 Massive amount of memory and parallel processing power for under $1M 1 server...8 CPUs, each CPU with 8 cores for $100,000 can execute 64 threads in parallel A A A A A A A A A A A A A A A A can hold memory modules adding up to ~2TB today 100GB/sec data throughput per server 8 blades...for less than $1M can execute 512 threads in parallel A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A can hold memory modules adding up to ~16TB 2011 SAP AG. All rights reserved. 4

5 Simplified Memory Hierarchy (based on Intel Nehalem) CPU Core Core Core Core Size Latency 1 st Level Cache 2 nd Level Cache 1 st Level Cache 2 nd Level Cache 1 st Level Cache 2 nd Level Cache Shared 3 rd Level Cache 1 st Level Cache 2 nd Level Cache 64KB 256KB 8MB ~4 cycles 2 ns ~10 cycles 5 ns cycles 20 ns Main Memory several GBs up to TBs ~ 200 cycles ~ 100ns Disk several TB several million cycles several ms 2011 SAP AG. All rights reserved. 5

6 Main Memory Bottleneck Disk is Tape 10 5 times slower compared to main memory access No disk access during normal operation Use disk as kind of archive CPU Core CPU Cache CPU Cache MainMemory Performance bottleneck today: CPU waiting for data to be loaded from memory into cache Memory Price / MB AM is going down 64 bit architectures make maximum limit of 4GB main memory obsolete Disk Performance bottleneck in the past: Disk I/O Memory Access Access to main memory is not arbitrarily fast Memory bandwidth increases but memory access latency remains the same Cache misses limit performance Traditional database: CPU spends 50% of time waiting after cache misses (1999) Transfer memory-cache is block wise (cache lines), for example 64 bytes Main memory access benefits from data locality, random access is slow 2011 SAP AG. All rights reserved. 6

7 In-Memory Computing In-Memory Computing Keep all required data in main memory arely accessed data can be moved to disk (e.g. after 1 year) Compress data in memory Disk I/O no longer an optimization target Cache sensitive data layout High locality (data that is needed together is stored together) Compression (decompress in cache) Parallelization Data structures allow splitting into pieces that can be processed in parallel Avoid locks Unify OLAP and OLTP systems : combinte column and row store Column store for reporting like access ow store for OLTP like access 2011 SAP AG. All rights reserved. 7

8 ow Based and Column Based Data Storage Country Product Sales US US US JP UK Alpha Beta Alpha Alpha ow-based Aggregate sales figures: andom access! US Alpha US Beta JP Alpha 700 UK Alpha 450 Sum(sales): Sales: 4 byte, country: 2 byte, product: 10 byte Cache line: 64 Byte 48 byte waste = 75% Columnar US US JP UK Alpha Beta Alpha Alpha data stored in contiguous memory High locality 2011 SAP AG. All rights reserved. 8

9 ow Based and Column Based Data Storage? Column Store Many calculations on single or few columns only Searches based on values of a few columns Big number of columns Big number of rows and columnar operations are required aggregate, scan, etc. High compression rates possible Most columns contain only few distinct values ow Store Application often needs to process single records at one time many selects and /or updates of single records Application typically needs to access the complete record Columns contain mainly distinct values Aggregations and fast searching not required Small number of rows (e.g. configuration tables) 2011 SAP AG. All rights reserved. 9

10 SAP HANA database Overall Architecture of the engine NewDB Clients (Application, Analytics Technology, etc) NewDB Connection and Session Management Session Parameters equest Parser SQL Script MDX Planning Engine equest Processing And Execution Control Authorization Optimizer Calc Engine Metadata elational Engines In Memory ow Store Execution Layer In Memory Column Store Disk Based Store Temporary esults In Memory Object Store Transaction Data Aging Persistence Layer Page Management Logger Data Volumes Disk Storage Transaction Log Volumes 2011 SAP AG. All rights reserved. 10

11 ow Store Architecture ecord Level Write Locks Transaction Execution Layer ow Store ow Store Transactional Version Memory owid ecord Versions Index Write Operations ead Operations Index key rowid Intermediate esults Segment Page Header Slot Page Data records managed in pages Temporary versions in separate memory Index only in memory Optimistic latch free index access Loads all tables at startup Incremental transaction logs Parallel write and restart Data persisted during savepoint operation Free Pages, Sparse Pages Page owid ecord Versions Version Memory Consolidation Checkpoint Writer Page Log eplay/ Undo Agent Write log entries invoke checkpoint Write pages Persistence Layer 2011 SAP AG. All rights reserved. 11

12 Column Store Architecture SQL Processor Calc Engine Query (search) Column Store Optimizer Write equest Column Store Column Store Own optimizer and execution control OLAP Optimizer OLAP engine: Special optimizer and operators Column Store Execution Plan Column Store Executor For each column ow Level Locks Compressed main storage Standard Operators Operators optimized for OLAP Column Store Operators (ead) Delta storage with only basic compression Write Operations Write operations go into delta only History Storage Main Storage Delta Merge Operation Delta Storage Consistent View Filter Consistent View Transaction Merge operation re-encodes main area Optional history storage (also main + delta) Delta log written per table Main Delta Load/Unload Agent Per Transaction Change Information Main area persisted during merge operation Write main and delta log virtual files Persistence Layer Load table Write delta log Transaction Status Write and read possible during merge Persistence Interface Logger Loads tables on demand (preload configurable) Virtual Main Files Virtual Delta Log Files edo and Undo Log 2011 SAP AG. All rights reserved. 12

13 Persistence Layer NewDB In-Memory Stores Log eplay/ Undo Agent Page Checkpoint Writer ow Store ow Store Pages Columns in contigous memeory Column Store Load/ Unload Agent Transaction Allocate memory callback Same Memory Location Page Management Persistence Interface Absolute Page API Virtual File API Free Block Page Status Info Consistent Change Lock Undo Converter Converter Table Cache Externally Allocated Data Pages Data Cache Other Data Pages Data Pages of Virtual Files Page Buffer Log Queue edo Log Differential Log Logger Save Point Coordinator Page I/O Disk Storage Data Volumes estart ecord List of open transactions Persisted Converter Table Physical Data Pages Transaction Specific Undo Containers Other Data edo edo Log Log Log Volumes Differential Log estart Agent edo lists Persistence Layer 2011 SAP AG. All rights reserved. 13

14 SAP HANA database - Assets High Performance and Scalability Compression, cache-aware storage Executing application logic inside database Distribution educing Complexity and Cost Hybrid: column based, row based (future: object store and disk based) in one system Analytics and OLTP in one system Support for OLAP MDX, calculation engine, OLAP views Compatibility and Standard DBMS Features SQL, ACID run SAP applications that use Open SQL without changes. Multi Tenant Support Support For Temporal Tables time travel, time slider applications, reporting using historical data, change recording Main memory text search engine 2011 SAP AG. All rights reserved. 14

15 SAP HANA database A Distributed Database Different Options Data Distribution No Tenant Separation Host Host Host NewDB System Database Process Database Process Database Process Data Partition Data Partition Data Partition Multi Tenant System Host Host Host Tenant 1 Tenant 2 Tenant 3 Tenant 4 Tenant 5 Tenant N NewDB System Database Process Database Process Database Process Database Process Database Process Database Process Tenant Data Tenant Data Tenant Data Tenant Data Tenant Data Tenant Data Default: tenants are not distributed across multiple servers 2011 SAP AG. All rights reserved. 15

16 SAP HANA database A Distributed Database Topology Information / metadata Host Slave Name NewDB Database Topology and distribution Information (replicated) Fast read access via shared memory Topology And distribution Information Master Name Host Slave Name NewDB Database Topology and distribution Information (replicated) Host Slave Name NewDB Database Topology and distribution Information (replicated) 2011 SAP AG. All rights reserved. 16

17 Tenant Separation Table T1 T1 in Tenant 100 T1 in Tenant 200 T1 in Tenant 300 Client A B C A B C A B C A B C y c 765 y c 770 z a 733 x b z a 701 w c 800 z b 777 z a w c x b z a z b defined by central metadata ABAP client concept NewDB tenant concept NewDB tenant Separate database process Separate data volumes Separate instances of tenant-dependent tables Separate transaction domain 2011 SAP AG. All rights reserved. 17

18 Temporal Tables Temporal Table (Logical View) ow ID Text Size Shirt XL Shoe M Hat L ow ID Text Size Shirt XL Shoe M Hat L Valid-From : : :11 Valid-To UPDATE article SET SIZE = S WHEE ID= 712 Valid-From : : :11 Valid-To : Hat S :05 Currently only in column store historical insert ow ID Text Size Shirt XL Shoe M Hat L DELETE FOM article WHEE ID= 546 Valid-From : : :11 Valid-To : : Hat S :05 historical 2011 SAP AG. All rights reserved. 18

19 In-Memory Computing Business Database Move data-intensive operations to the data layer Do the relational operations at database level ( i.e. select for all entries ) Hierarchy handling ( i.e H cost center hierarchy ) eporting and planning functionality Planning engine Transactional behavior Unify DB and application transaction Push snapshot isolation to the database / remove transactional buffer in the appserver layer Business Function Library Provide business functions in the DB layer like Currency /Unit conversion Date / time /fiscal period /calendar calculation Statistical functionality Deep integration with the application server Fast communication layer SQL extensions ( SQL script ) New data types ( text, GUID, ) 2011 SAP AG. All rights reserved. 19

20 Thank You! Questions? Jürgen Primsch SAP AG