Exadata High Volume Testing Customer POC Findings - Himesh Bhatt

986 views
889 views

Published on

Published in: Technology, Business
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
986
On SlideShare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Exadata High Volume Testing Customer POC Findings - Himesh Bhatt

  1. 1. <Insert Picture Here>Exadata High Volume TestingCustomer POC FindingsHimesh BhattOracle EMEA Infrastructure Solutions
  2. 2. Agenda • Customer Overview • Decision for the POC • POC Preparation and Execution • Results and Achievements • Volume Tests • High Availability Features • Lessons Learned • Conclusion • Q&ACopyright © 2010, Oracle Corporation and/or its affiliates –2–
  3. 3. Customer Overview • Business • B2B retailers selling office supplies and stationary • Operate in over 30 countries • Expected business growth ~30% within next 4 years • Challenges • Invoicing – Long batch run times, unable to process invoices daily • High costs associated with administering many large SAP databases • New business B2B functionality could not be implemented on existing infrastructure • Decided to renew infrastructure to meet future business requirements • Oracle solution: • Exadata for performance, consolidation and business agility • Application instances on x86 running Solaris and Industry SolutionsCopyright © 2011, Oracle Corporation and/or its affiliates –3–
  4. 4. Decision for the POC • Stated Goals • Expose Exadata to the customers SAP business critical processes • Simulate real life customer SAP workload on Exadata • Measure performance improvements • Throughput • Response times • Demonstrate how Oracle technology can minimize business disruption caused by hard or soft failures. • Ground Rules / Conditions • No tuning or tweaking of database • Use standard Exadata deployment without optimization • Customers IT departments should be able to easily repeat all improvements on their production systemsCopyright © 2011, Oracle Corporation and/or its affiliates –4–
  5. 5. Conducting the POC • Preparation – 3 Weeks • Identification and validation of real live test cases • Simulation of online users in a web shop • Batch jobs and customer invoicing • Design, build and install entire landscape for SAP • Creation and load of customers database • 6.5 TB DB migration from Oracle 10.2 (SPARC) via Oracle 11.2 (x86 Linux) into an Exadata ASM/RAC environment • POC timelines ( 1 Week) • On-site activities at Oracle OSC – Linlithgow, Scotland • Load injection • Deployed load runner and Control-M • Executed remotely from customers siteCopyright © 2011, Oracle Corporation and/or its affiliates –5–
  6. 6. Overview of POC Infrastructure Excluding servers for load generation and application tier for JAVA web shopCopyright © 2011, Oracle Corporation and/or its affiliates –6–
  7. 7. Result: Volume Testing - Batch • Invoicing of Sales Subsidiaries • Job scheduler starts 900+ batch jobs • At any given time, between 1 to 130 jobs are being executed in parallel • Many jobs with pre-defined dependencies • Runtime observations • Similar workload on customer existing environment takes 11.30 hrs to complete • Completion time on Exadata 3.45 hrs • Two active nodes on a Half Rack Exadata had been utilized to 25% of capacity • Customers workload completed in one third of the time on Exadata, compared to the customer existing SAP environmentCopyright © 2011, Oracle Corporation and/or its affiliates –7–
  8. 8. POC CPU Utilization Mixed workload: single and parallel runs with many SAP programsCopyright © 2011, Oracle Corporation and/or its affiliates –8–
  9. 9. Exadata Impact on Business Responsiveness • Response time reduced significantly: 3.45 Hrs .. but • Newer Intel CPU technology had been deployed in the POC, in comparison to the customers exiting environment • Rerun test using DB staging environment (no Exdata usage) • Runtime using single instance DB server was 7.20 Hrs • Observation: • Runtime 11:30 Hrs → 7.20 Hrs • Deploying faster processors (App and DB Tier) and upgrading storage reduced runtime from 11.30 Hrs to 7.20 Hrs • Runtime 7.20 Hrs – 3.45 Hrs • Significant improvements when deploying Exadata technology (Flash, Infiniband) and Exadata S/W (Smart Flash Cache ..)Copyright © 2011, Oracle Corporation and/or its affiliates –9–
  10. 10. Results Volume Testing – Number of Online Users • Simulation of online orders entered in a web shop • Runs with various parameters ( #users, thinktime, # RAC Nodes) • After the initial throughput was achieved, scalability within the SAP application had been identified and eliminated • Throughput • One node RAC: 7500 users with 4s think time: • => ~1s response time and 64% CPU utilization (one node) • Four node RAC: 6000 users with 4s think time: • ~0.9s response time and 17% CPU utilization (all nodes) • Exadata over achieved the goal of 2500 users and still had a lot of spare capacity.Copyright © 2011, Oracle Corporation and/or its affiliates –10–
  11. 11. High Availability Features Mixed workload: single and parallel runs with many SAP programsCopyright © 2011, Oracle Corporation and/or its affiliates –11–
  12. 12. High Availability Features Derived from /nST03, DBA Cockpit & AWR Reports Key Figure / Per Batch Online Second [13% util of POC [17% util of POC hardware] hardware] User Calls 39.838 69.285 Physical Writes 6.538 2.952 Physical Reads 14.727 4.404 Logical Reads 348.300 389.741Copyright © 2011, Oracle Corporation and/or its affiliates –12–
  13. 13. High Availability Features of Exdata • Shutdown RAC nodes during normal SAP activity • Show case automatic reconnection from SAP Application Servers to RAC database node. (TAF) Transparent Application Failover • Demonstrate significant reduction of unplanned downtimes in comparison to traditional HA/Cluster solutions. - Remove need for Oracle Database instance recovery • SAP Replicated Enqueue with Sun Cluster • Demonstrate a solution which ensures continuous business operations. • Example case: Losing the SAP central lock table held in main memoryCopyright © 2011, Oracle Corporation and/or its affiliates –13–
  14. 14. SAP Instance Reconnect Transparent Application Failover Shutdown Abort RAC NodesCopyright © 2011, Oracle Corporation and/or its affiliates –14–
  15. 15. Lessons Learned • Exadata provides better performance – out-of-the-box • You will see instance performance benefits • No Exadataor DB specific tweaking/optimization was needed • However, effect of new technology should be considered as well when making comarisons (e.g. faster CPU’s on App Servers) • We were not able to saturate Exadata • No bottleneck in the hardware/infrastructure were seen or experienced • Throughput was limited by contention within the application: • Assignment of new documents numbers for new postings in SAP application. • Hot spot inserts at the end of the tables (ascending primary key) • In order to achieve even higher throughput numbers and take full advantage of Exadata you need to resolve these bottlenecksCopyright © 2011, Oracle Corporation and/or its affiliates –15–
  16. 16. Lessons Learned – continued 1 • User I/O is no longer a major wait event • Cell single block physical read – extremely fast • Similar action to a db file sequential read in a non-Exadata environment • Oracle RAC with SAP on Exadata Scales!! • Almost same throughput numbers in 1 – 2 – 4 node RAC configurations • Slightly higher CPU consumption in n-node RAC configurations • Collecting DB Workload & Utilization Data for Sizing • During test runs AWR reports have been captured along with Exadata utilization data • Load profile can now be compared with DB workload from other customers for sizing & migration considerationsCopyright © 2011, Oracle Corporation and/or its affiliates –16–
  17. 17. Lessons Learned – continued 3 • No Smart Scans in these business-critical workloads • Well tuned application setup and DB-index design: • unique index access • selective secondary indices • no tablespace / index scansCopyright © 2011, Oracle Corporation and/or its affiliates –17–
  18. 18. <Insert Picture Here>SAP Applications with Oracle Exadata DatabasePrerequisites and Deployment OptionsHimesh BhattOracle EMEA Infrastructure Solutions

×