Enter Date in Title Master Workload Management HBC Case Study IRMAC, January 2008 Shelley Perrior -DBA team lead
Agenda Agenda: Corporate and presenter background WLM case study System profile Business goals WLM configuration Other points
About HBC About HBC: Hbc – the Hudson’s Bay Company Canada’s oldest and largest diversified general merchandise retailer Incorporated on May 2, 1670 More than 580 Stores in all 10 Provinces across Canada 5 Retail formats, along with an eRetail portal The Bay – Department Zellers – Mass Home Outfitters – Specialty home Designer Depot – Deals outlet Fields – Discount 70,000 associates focused on exceptional customer service Privately owned as of 2006
Database Group - HBC Presenter Background: Supervisor database administration for all databases Teradata technical lead, Teradata Certified master Liaison with business and application teams Extensive experience in system and application performance tuning on Teradata
Configuration Of Data Warehouse: Database is Teradata V2R6.1.1 BI reporting tool is Microstrategy 8.0 Current production system is a 12 node 5380 system with 7.5 TB of data Dev system is a 4 node 5350 system. Both Prod and Dev databases are at V2R6.1.1 Current workload consists mainly of a Business Intelligence application using Microstrategy for reporting and our Fraud Control application. Also have limited adhocs running as well. We use TDWM and Priority scheduler to monitor and control workload.
Business goals to address with WLM Goals in priority order: Goal 1: Ensure a response time of less than 10 seconds from the POS for Returns transaction validation Goal 2: Ensure 95% of all Microstrategy queries finish in under 10 minutes Goal 3: Ensure ETL processing completes within specified batch window Goal 4: Accommodate effectively, emergency requests and changes in workload (daily, monthly, weekly etc)
Problems if WLM not used: Heavy workload from Microstrategy can threaten the response times for the higher priority work. Potential runaway queries may monopolize system resources Changes in workload may not be handled effectively (i.e.. batch runs too long and impacts business reporting)
WLM Configuration: Work categories (with different WLM controls) Transaction load and verification from POS Microstrategy Reporting ETL Other WLM controls: Resource allocation & priority controls CPU allocation: Ensure a percentage reserved for high priority work (expedited workload) Use milestones to push short queries through faster Spool space usage limits Allocations change by time of day (for ETL) controls set up by user account code
WLM Configuration cont’d: Individual user request controls Max allowable spool priority of query by user account code Concurrency controls Limit number of reports per individual, total number on system, no batch in certain windows. Emergency Query privileges Allow for special situations when queries may need a higher priority.
Other points: WLM implementation process: Work with application to identify business requirements Identify query grouping (i.e.. application, type of query, etc) Collect and analyze data on resource usage implement WLM controls and monitor adjust as required until desired results are met Set up SLA set up monitoring and alerts
Lessons Learned: -Workload management is an ongoing process -Good communication is necessary between applications and database team -SLA’s are necessary -Monitoring critical -Historical data important