Orange County Convention Center Orlando, Florida June 3-5, 2014 3807 - Lockheed Martin takes Grouping, Pegging & Distribution (GPD) for a Test Drive on HANA Alan Atkins, Lockheed Martin Jeff Morin, Lockheed Martin
Agenda Learning Points Lockheed Martin Company Info HANA The Basics HANA Testing and GPD @ Lockheed Martin Challenges Best Practices Key Learnings
LEARNING POINTS Overview of Lockheed Martin's Space Systems GPD solution Review of Lockheed Martin's experience with ECC on HANA Lockheed Martin's approach & results from testing GPD as part of ECC on HANA
Lockheed Martin Company Overview 2013 Revenue - $45.0B
Our SAP Functional Baseline Mission Missiles & Space Info Sys & Aero MST MFC Space IS&GS Sys & Train Fire Ctrl Systems Global Sol Enterprise Ops Finance - Operations - - Procure / Pay Business Analytics Intelligence - HR/Learning With 120K end users and 35K professional users, SAP is widely used
HANA 101 HANA is an In Memory Database appliance Combining Hardware and Software Innovations to Accelerate Business!
SAP HANA Primary Use Cases ABAP Accelerator / Side Car Analytics Business Objects BOBJ / Other Clients BW BOBJ / Other Clients OLTP Business Suite Business Suite Business Suite BW Business Suite AnyDB HANA AnyDB HANA HANA HANA SLT SLT In Use @ LM? Business Suite = ECC, SCM, SRM, or CRM
ECC on HANA Implementation Considerations Requires Specialized HANA appliance hardware Ehp 7 Required System must be Unicode Existing databases converted to HANA using OS/DB migration Z-Code Remediation A small percentage of programs must be fixed in order to guarantee functional correctness under HANA Other programs need to be changed to meet existing performance, or experience further optimizations on HANA SAP supplies tools, but does not identify all issues. Significant testing required.
LM Space Systems GPD Fun Facts Running GPD Since 1999 Single Level Grouping Only (Type 1) 20 Plants No Stock Transport Orders (STO) Custom Excess, Scrap & Loss logic Total Pegging Table Size approx 2M records Using RDISALL (old EP4 distribution) Run 5 Variants of Distribution in Production Weekly No Cross-Company Code Distribution
ECC on HANA CUV Process at LMCO SAP s Customer Validation (CUV) Program is - A Proof of Concept Before Ramp-Up / General Availability Free of Charge With Direct Development Support SAP provides - Support for Installation of Software CUV Specific Support Packages On-going Support for Issues Technical and Functional Ends with Scorecard for feedback Ran 4 passes of testing from service pack 1 (April 2013) through 4 (May 2014) RESULTS: Very positive! SAP resolved issues in hours/days. Direct line to OSS and GPD Developer Resulted in a number of fixes for the next General Release support packs.
Overall ECC on HANA Testing Objectives Objective 1: MRP/GPD performance testing (HANA) Compare Run Times Baseline Non-HANA to HANA environment Objective 2: Long running jobs performance testing Compare run-time of key long running batch jobs Objective 3: ECC General Baseline Testing Validate No Changes for GPD in Ehp7 Use MRP / GPD baseline on ECC 6/ Ehp5 (Production) Run GPD and Compare results to production Cost Shift Analysis Objective 4: Transaction Testing (Optimized) Review list of optimized SAP transactions, and execute long running t-codes and compare time between both systems. Objective 5: Standard Most Used Transactions Review list of most used transactions. Execute and compare between systems.
LM Space Systems HANA Testing Landscape Source System RHEL/Oracle ECC 6.05 Production Unicode Mitigations Baseline captured from production after full weekend process completed App Server RHEL DB Server RHEL-Oracle System Copy - Export ECC 6.05 Test Baseline RHEL/Oracle ECC 6.17 Production copy imported System upgraded to ECC 6, Ehp 7 Support Packs applied General Release CUV specific support packs applied (HANA specific) App Server RHEL DB Server RHEL-Oracle Baseline System (ORACLE) System Copy - Import Upgraded to ECC 6.17 Net Weaver 7.40 System Copy - Export ECC 6.05 ECC 6.17 Test System RHEL/HANA ECC 6.17 Baseline system copied to HANA system for testing App Server RHEL DB Server IBM Appliance- HANA Evaluation System (HANA DB) System Copy Import (Start Testing) ECC 6.17
GPD Specific Testing Scope Functional Testing Individual transaction testing, ensuring major variations of functionality worked ( Sunny Day ) Cost Shift testing Full GPD testing, comparing HANA GPD results to base lined pre-hana data, ensuring no undesired cost shift. GPD Based Custom Reports Key reports using large amounts of data Pegging reports Commitment Reports Weekly Error Checking Reports Performance Testing Intent to compare run times between similar HANA and Non-HANA environments. Ran full weekend process (MRP, GPD, Reports) in both environments, comparing times GPD Based Custom Reports Ran key reports using large amounts of data
GPD Functional Results - Findings Early Challenges Many Bugs (EHP7, SP1-SP2) Pegging log fills up to 999999 errors, dumped. Required OSS note: 1965775 - Correction for PEG01N Optimization. Calls to HANA Stored Procedures commented out. Various Technical bugs Database procedure proxies, time zone bug, daylight saving time bug, etc. Database connection issues Stored Procedures bugs - Pegging procedures executed at the database level generated a number problems. The fix via SAP (CUV process) was often to remove these procedures from the database level and implement back at the app level. 1973247 - Correction for PEG01N on HANA : Fetch on PEG_TASS to return custom field information Did not accommodate z-fields in peg_tass 1891842 - Rounding issues with the HANA stored procedures for the PEG01N transaction Caused over pegs resulting in undistributed replenishments 1965775 - Correction for PEG01N Optimization - Relevant to EHP7
GPD Functional Results - Findings Custom Reports Not optimized for HANA, some ran significantly longer. GPD Commitment report Ran 1000x longer, due to nested select statement in ABAP code. Custom Pegging (peg_tass) report Ran 10x longer, due to bug in Group reading function module. Applied Note 1904492 - GPD616+: Optimized MRP group read in GRPG find modules Undistributed Group WBS Report Short dumping due method of database call Revised OBJNR_PACKAGES parameter in report. Financials Interface Various bugs causing perpetual run time. Various GPD past fixes (OSS Notes), were not migrated to HANA version, required re-application
GPD Functional Results Cost Shift Performed a Before/After results analysis of GPD Results Ran GPD in Production, saved Baseline Ran GPD in Oracle Baseline Client (ECC 6.17-only impact) Ran GPD in HANA Client Common Oracle & HANA Pegging/Cost shift 655 pegs HANA-only additional shifts 140 pegs Note: Total Pegging Records 2,037,314 Amounted to $96K additional unpegged cost (.0014% of cost) and some movement between WBS Elements Cost Shift considered negligible
GPD Performance Results - General Run Time Results - ECC 6.17, SP4 HANA 7000 SECONDS 6000 5757 5000 4391 4,414 4000 3000 2000 1000 2267 1378 2193 1079 3,137 0 MRP PEGGING DISTRIBUTION GPD REPORTS ORACLE HANA 39% faster 62% faster 75% faster 29% faster GPD Run Time 60% Faster with HANA!
GPD Performance Results GPD Reports 1800 1698 SECONDS 1600 1400 1241 1200 1086 1015 1000 802 800 605 600 497 400 Key GPD Intensive Reports tested, all custom reports No optimizations made to reports Speed improvement found in most reports OVERALL: 29% Faster on HANA 200 0 156 21 18 45 47 80.17 20.2 2.7 0.87 126.38 69.87 ORACLE HANA
Performance Results Other Transactions Transaction Results - ECC 6.17, SP1 HANA Individual transaction test included : Z-Transactions Optimized for HANA transactions All but one transaction had performance improvement Tested Online Transactions 60% Faster on HANA!
Performance Results Aero Current Production Environment 50000 SECONDS 45000 46980 Pegging Table Records MRP, weekly 12.8 million 3 hours 40000 35000 34596 Pegging, weekly Distribution, weekly 9 hours 9 hours 30000 25000 26100 Results are from EHP7/SP3 Non-similar hardware compared Significant hurdles with stored procedures for pegging (resolved) 20000 15000 10000 5000 0 10476 5100 1530 MRP Pegging Distribution Oracle HANA 70% faster 24% faster 78% faster Combine MRP/GPD 60% Faster on HANA
Performance Results Space FICO Current Production Environment CO records (COEP) 741 million Assessments, weekly 4.5 hours Settlement, weekly 2 hours Billing, weekly 6 hours Results are from EHP6/SP2 Non-similar hardware compared Mixed results a blend of z-code issues and DB insert speeds Early HANA version with improvements forthcoming. 1600 Extrapolated Hrs/Year 1400 1373 1200 1112 1000 844 800 600 400 284 200 65 48 46 66 0 Allocations Settlements A&D Billing Reports/Extracts Oracle HANA 26% faster 42% slower 19% faster 66% faster SP2 was 31% faster overall than SP1
Challenges Be wary of custom fields in core tables Found some HANA stored procedures that did not read z-fields, resulting in errors e.g PEG_TASS table Baseline setting Apples to Apples Need consistent baseline / HANA boxes to compare results Similar hardware, memory, OS, configuration/parameter settings. Caching GPD results affected by caching, need to ensure caches are cleared before performing subsequent runs.
Key Learnings / Best Practices CUV Installation - Save a general release baseline when planning to test multiple releases CUV Service Packs (SP) cannot be upgraded Backup last general release SP before proceeding to CUV service packs CUV Process Exceeded expectations Plan significant amount of technical (BASIS) time for installation/upgrade. The majority of our time was spent here. New Technology, unexpected difficulties. SE16 Fast!
Key Learnings / Best Practices OSS First stop when encountering unexpected results as historical notes may be inadvertently omitted from HANA releases. Custom Reports - Spend time looking at customized reports At a minimum remove problems using SAP provided code analysis tools Optimize reports to take advantage of acceleration Don t always trust the optimizer tools. They are only guidelines. We are currently analyzing over 700 reports/custom programs Newer analysis tools available, worth investigating Expanded Tools: Clone-Finder, Code Inspector, ABAP Test Cockpit, Custom Code Lifecycle Management New Tools: ABAP Procedure Monitoring, ABAP SQL Monitoring and SQL Performance Tuning Work List (SWLT) Invest
FOLLOW US
THANK YOU THANK YOU FOR PARTICIPATING Please provide feedback on this session by completing a short survey via the event mobile application. SESSION CODE: 3807 For ongoing education on this area of focus, visit www.asug.com