Presentation on theme: "Computer Measurement Group, India 1 1 www.cmgindia.org Workload Model in Airline Operations Manoj Raghavendra, TCS."— Presentation transcript:
Computer Measurement Group, India 1 1 Workload Model in Airline Operations Manoj Raghavendra, TCS
Computer Measurement Group, India 2 Contents Introduction of the System Load Model – Scenarios & Approach Key Benefits – Capacity Planning – RAC Configuration
Computer Measurement Group, India 3 EDA Cache for CM means A system for publishing real-time events A system to publish timed passenger summary event A Business data cache that can be substitute back-end calls for a subset of Passenger and Flight data The System XML DB Amedeus Global Travel Ops System Amedeus Global Travel Ops System EDA Components contributing to workload Harvesters – 1A and 1B pattern Data Enrichers (1B Pattern) Data Manager (1D, 1E and 1F patterns) Event Managers – 1A, 1C Pattern EDA Other Consumer support systems Flight info, Boarding, Conformance, Disruption Flight info, Boarding, Conformance, Disruption CM 1C, 1D 1A 1D,1E 1B 1F EDA EDA Load model was proposed to estimate the saving in the number of calls to Amedeus EDA NFRs 1200 flight departures per day 200,000 pax per day 180 avg pax per flight 10% growth year-on-year SLA ms for Harvesters 100 ms for Event Managers 400 ms for Data managers Tech Stack: Auria Sonic ESB, Sonic MQ, DataXtend Semantic Integrator, Oracle11g XML DB
Computer Measurement Group, India 4 The Load Model Production Data collected: Selected the Peak day of the year (15- July -2013): 1222 flights & 190,000 pax Selected a Disruption day (Volcanic Ash Cloud): 486 flights rescheduled 350 flights cancelled 80 flights rerouted before dep 30 flights rerouted after dep Data available at Business activity level Creation of Load Model Patterns defined for each scenario Mapped against the Business activity Production data applied on the Pattern - Business activity Service level throughput computed - using an custom app
Computer Measurement Group, India 5 Key Benefits – RoI & Capacity sizing Calls to Amedeus reduced from 170 TPS to 150 TPS Derived the Service level throughput on the XML DB (CRUD ops) ~ 300 TPS New Requirement – Add Recovery of EDA Cache –Additional throughput of ~400TPS for 60 minutes to recover imminent flight details –Additional throughput of ~100 TPS for 6 hours to recover D+3 days flights CPU Sizing of the Oracle XML DB ~ 730 TPS at DB level – 100 Cores of HP G7 processor With Recovery, additional ~400 TPS on the DB – 152 cores of HP G7 processor Additional CM Applications opted to use EDA Cache Load on Amedeus reduced further to 140 TPS Increased the load on EDA DB ~ 900 TPS -200 cores of HP G7 Load Model was effective in assessing the RoI and Decision making with changing requirements
Computer Measurement Group, India 6 Key Benefits - How much to RAC Oracle XML DB - # Cores required 100 cores ~ 4 Node RAC (32 G7 core per node) 180 cores ~ 6 Node RAC 200 cores ~ 8 Node RAC Performance test results showed RAC nodes could not scale linearly for EDA – Prime workload (>10 tps) derived from Load Model constituted 80% of Workload G8 processors are 1.5 times faster than G7 G8 v2 processor are 2.25 times faster than G7 Decision made to go for 2 node RAC with G8 v2 processors.
Computer Measurement Group, India 7 Questions ???