SAP HANA SPS 09 - What s New? SAP HANA Scalability

Transcription

2 Disclaimer This presentation outlines our general product direction and should not be relied on in making a purchase decision. This presentation is not subject to your license agreement or any other agreement with SAP. SAP has no obligation to pursue any course of business outlined in this presentation or to develop or release any functionality mentioned in this presentation. This presentation and SAP s strategy and possible future developments are subject to change and may be changed by SAP at any time for any reason without notice. This document is provided without a warranty of any kind, either express or implied, including but not limited to, the implied warranties of merchantability, fitness for a particular purpose, or non-infringement. SAP assumes no responsibility for errors or omissions in this document, except if such damages were caused by SAP intentionally or grossly negligent SAP SE or an SAP affiliate company. All rights reserved. Public 2

3 Scalability Setting the stage Scalability is the ability of a computer hardware/software to take full advantage of its changed context in a re-scaled situation, e.g. by adding or reducing resources Vertical Scalability (Scale-up) Adding resources within the same computing unit, e.g. #CPUs, #Ds Horizontal Scalability (Scale-out) Adding multiple computing units and making them work as one logical computing unit SAP HANA is designed for scale-up and scale-out since the beginning 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 3

4 Scale-up

5 Micro-Architecture Trends NUMA (Non Uniform Memory Access) remote CPU 1 CPU 1 local each socket has its own memory controller (IMC) Core1 Core2 Core1 Core2 scales well, but there are NUMA-effects Core3 Core4 remote Core3 Core4 memory access is non-uniform -> local vs. remote access high-speed interconnect CPU 1 CPU 1 Core1 Core2 Core1 Core2 Core3 Core4 Core3 Core SAP SE or an SAP affiliate company. All rights reserved. Public 5

6 HANA Parallel Query Processing multi-user multi-query ( multi-plan ) intra-plan-parallelism ( multi-pops ) intra-pop-parallelism ( multi-threads ) Core 1: Query 1: Join Aggregate Sort Core 2: Query 2: Join Aggregate Sort Core 3: Core 4: Query 3: Join Aggregate Core 5: Core 6: Jobs over time HANA global optimizer generates efficient execution plan for each query (rule-based, cost-based) Execution plans are cached and reused by HANA Each execution plan is split into multiple plan operators (e.g. join, agg, sort) that are executed in parallel Plan operators generate job graphs, containing job nodes Job nodes are dispatched for execution to available treads by the Job Executor 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 6

7 Summary: Vertical Scalability in HANA (scale-up) Parallelism in HANA HANA is heavily optimized for state-of-the-art HW architectures HANA uses massive intra-plan and intra-operator parallelism for query execution HANA tries to use all available resources for maximal parallelization Challenge: NUMA-effects on large multi-socket systems Remark: Virtualization of CPU resources can reinforce NUMA-effects Job Scheduler Central place in HANA to execute plan-operator jobs Re-adjust dynamically the concurrency level avoiding over-parallelization Can consider additional.ini configuration data to limit the concurrency level 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 7

8 Scale-out

9 Horizontal Scalability (scale-out) General Scale-out means expanding to multiple servers rather than a single, bigger server in one database system Multiple servers (nodes) are switched together to one logical, but physically distributed database system When to scale-out? Overcome hardware limitations of one single server (memory, cpu) Scale-out aims to distribute data and workload Attention Scale-out requires deep knowledge about data, application and hardware Scale-out increases data center operation costs Scale-out affects your scaling factor by inter-node communication 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 9

10 HANA s Shared-Nothing Architecture A shared-nothing architecture helps to minimize the inter-node communication in distributed environments Topology Each node runs on its own local data (shared-nothing) Data Volume Master node HDB net Standby node(s) without own persistence Log Volume Shared file system for node fail-over and recovery (HW partner implementation) Nodes communicate via internal HDB net protocol All nodes belonging to the same HANA system must have the same HW setup Data Volume Log Volume Slave node Standby node(s) Auto fail-over 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 10

11 Minimize Inter-Node Communication with Data Distribution Data distribution aims to find the best balance between data model, data usage and actual workload to minimize internode communication Analyze Distribute Data model Types of data, e.g. master data, transactional data and configuration data Referential integrity between tables Data usage Data interdependencies, e.g. table groups Data access paths Workload Mixed workloads, i.e. read/ write, analytical/transactional Update frequency Table co-location Locate interdependent tables on the same node, e.g. for referential integrity Locate tables within the same statement on the same node, e.g. for collocated joins or transactions Horizontal Partitioning Locate interdependent table partitions on the same node Table Replication Duplicate tables to multiple nodes, e.g. for master data referenced by many tables 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 11

12 HANA Data Distribution Table Placement Table Placement is the definition of generic table distribution patterns to control the physical location of (column-)tables and how many partitions of a table are to be created in a distributed HANA system Step 1. Classification In HANA it is possible to classify a table by additional metadata, like schema_name, group name, group_type The table classification can be set with the create / alter table statement The table classifications are stored in the table sys.table_groups_ Step 2. Table Placement Configuration For each table classification you can assign a table placement configuration A table placement configuration defines, on which node type, disk-volume (SPS09) a table and its partitions shall be located A table placement configuration also defines when a table shall be partitioned (works only if the table has a partition specification), e.g. initial_partitions, min_rows_partitioning and repartitioning_threshold The table placement values are inserted via SQL commands into table _sys_rt.table_placement (HANA studio -> SQL-console) When is the Table Placement Configuration evaluated During create/alter table and table redistribute operations The table placement can be overruled with the location_clause, partition_clause or move_clause information in the DDL statement 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 12

13 HANA Data Distribution Table Redistribution Save and restore old table locations Typical operations use a dedicated user with required privileges 1. Generate Plan Review Plan 2. Execute Plan Collects information about landscape and database Evaluates table placement information and.ini-parameter Apply heuristic algorithm to optimize table placement Save execution plan with execution order SYS.REORG_PLAN_ Read plan Parallel execution of plan items Plan is always executed completely No automatic rollback, but restore of old table locations possible Execute Plan 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 13

14 HANA Scale-out Solutions BW on HANA BW workload characteristics Mostly OLAP workload in rather static distribution environment OLAP load benefits from parallel processing on distributed partitions Master Node Handles OLTP load and DDL statements: ABAP system tables, meta data, operational tables and all row tables are stored Symmetric scale-out across the system Master-Node OLTP tables Slave-Node Slave Nodes will handle OLAP load exclusively: Master data + cubes/dsos/psas are distributed evenly across all slaves Cube, DSO and PSA tables are partitioned dependent on the table placement rules Useful information global.ini [table_placement] method= 2 (keeps row-store tables on the master node) note # (BW scale-out configuration recommendations) BW ABAP App-Server OLAP tables Slave-Node OLAP tables Standby-Node 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 14

15 HANA Scale-out Solutions Suite on HANA Symmetric hardware across the system Suite workload characteristics Mixed OLTP/OLAP workloads across all nodes (2PC, Cross-joins) Table access patterns dynamic and differ from customer to customer Dynamic Table distribution Buffered SQL statements in the statement cache are parsed and prioritized based on DB statistics Determine disjoint table groups Table groups are placed on same node while balancing memory and CPU Useful information note # (Suite scale-out configuration recommendations) Note # (Suite on distributed HANA database) Suite ABAP App-Server Master-Node Table groups Slave-Node Table groups Slave-Node Table groups Standby-Node 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 15

16 Takeaways Scale-up first! Scale-up Most common and easiest way for databases to scale No changes in the database or application required HANA parallelism strongly supports latest multi-cpu, multi-core HW architectures HANA NUMA-aware scheduling and memory allocation is under development Scale-out Requires deep knowledge about data and workload to achieve good scaling Scale-out works good in static distribution environments, like OLAP Scale-out in mixed OLAP/OLTP environments can be challenging Optimizing scale-out performance is an iterative task 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 16

17 How to find SAP HANA documentation on this topic? In addition to this learning material, you can find SAP HANA platform documentation on SAP Help Portal knowledge center at The knowledge centers are structured according to the product lifecycle: installation, security, administration, development: SAP HANA Platform SPS What s New Release Notes Installation Administration Development References Documentation sets for SAP HANA options can be found at SAP HANA Options SAP HANA Advanced Data Processing SAP HANA Dynamic Tiering SAP HANA Enterprise Information Management SAP HANA Predictive SAP HANA Real-Time Replication SAP HANA Smart Data Streaming SAP HANA Spatial 2014 SAP SE or an SAP affiliate company. All rights reserved. Public 17

19 2014 SAP SE or an SAP affiliate company. All rights reserved. No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP SE or an SAP affiliate company. SAP and other SAP products and services mentioned herein as well as their respective logos are trademarks or registered trademarks of SAP SE (or an SAP affiliate company) in Germany and other countries. Please see for additional trademark information and notices. Some software products marketed by SAP SE and its distributors contain proprietary software components of other software vendors. National product specifications may vary. These materials are provided by SAP SE or an SAP affiliate company for informational purposes only, without representation or warranty of any kind, and SAP SE or its affiliated companies shall not be liable for errors or omissions with respect to the materials. The only warranties for SAP SE or SAP affiliate company products and services are those that are set forth in the express warranty statements accompanying such products and services, if any. Nothing herein should be construed as constituting an additional warranty. In particular, SAP SE or its affiliated companies have no obligation to pursue any course of business outlined in this document or any related presentation, or to develop or release any functionality mentioned therein. This document, or any related presentation, and SAP SE s or its affiliated companies strategy and possible future developments, products, and/or platform directions and functionality are all subject to change and may be changed by SAP SE or its affiliated companies at any time for any reason without notice. The information in this document is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. All forwardlooking statements are subject to various risks and uncertainties that could cause actual results to differ materially from expectations. Readers are cautioned not to place undue reliance on these forward-looking statements, which speak only as of their dates, and they should not be relied upon in making purchasing decisions SAP SE or an SAP affiliate company. All rights reserved. Public 19