Transcription
PERFORMANCE ASSURANCEFOR BIG DATA APPLICATIONSBoris Zibitsker, PhDwww.beznext.combzibitsker@beznext.comSPEC Presentation
INTRODUCTIONwww.beznext.comBEZNext All Rights Reserved2
Scope Development of performance assurance technology, includingperformance engineering and capacity management and providingperformance assurance services Incorporation of advanced analytics including descriptive, diagnostic,predictive and prescriptive analytics during data collection, workloadcharacterization, performance evaluation, performance management,workload management and capacity planning during application,system and data life cycle Development Recommender, taking into considerationresponsiveness, availability and cost requirements Incorporation of Big Data capabilities for development enterpriseperformance assurance platformwww.beznext.comBEZNext All Rights Reserved3
PROBLEMwww.beznext.comBEZNext All Rights Reserved4
Problem from the BusinessProspective Business needs To make effective business decisions fast To increase profitability and reduce IT cost Business requirements to IT Activity of group of business users, customers and vendors using applications of the line ofbusiness is a workload Applications should: Provide Information necessary to support line of business decisions Answer specific What If business questions Generate prescriptions for how to make effective business decisions Each Workload has Service Level Goal (SLG) Responsiveness Demand for resources Data Each Line of Business has: Different budget limitations Different plan of development and implementation of new applications and modification of the existingapplications Different plans of growth and increase in volume of data and number of userswww.beznext.comBEZNext All Rights Reserved5
Problem from IT Prospective How to design and develop Big Data applications satisfying functionaland performance requirements (SLGs) of each line of business How to plan and cost-effectively manage Big Data infrastructure tomeet SLGs of each line of business How to set realistic expectations How to continuously and cost-effectively meet Service Level Goals foreach line of businesswww.beznext.comBEZNext All Rights Reserved6
Why Rate of Deployment of Big DataApplications is Slower than Expected Complex Technology Difficult to manage Security and privacy ApplicationsInterest in Big Data is high,so why is the rate of Big Dataapplications deploymentslower than expected? Use of advanced analytics Workload growth and new applicationslike IoT increase demand andcontention for resources People Difficult to hire experts Uncertainty and Risk of Surprises New applications New releases of software Workload management, performancemanagement and capacity planningwww.beznext.comBEZNext All Rights Reserved7
SOLUTIONS
BEZNext SolutionsPerformance Assurance for Big Data World Performance Engineering Capacity ManagementProcessUse of Advanced Analytics Descriptive Analytics Data Collection Diagnostic Analytics Workload Characterization Predictive Analytics Workload Forecasting Prescriptive Analytics Workload Management Performance Management Capacity Planning Verification Controlwww.beznext.comBEZNext All Rights Reserved9
Hadoop/YARN Data Operating System forBig Data Workloads Hadoop 2.x supports concurrentManagementDecisionsBusinessreal time, interactive and batchworkloads Complex multi-tier, distributed,virtualized, parallel processing,interdependent architectureBusinessAnalyticsReal TimeBatch YARN rules control clusterresource allocation, and mixworkload managementEDWData MartsYARN Data OperatingSystemDataReservoirData Lake Hadoop HDFSSources of Datawww.beznext.comBEZNext All Rights Reserved10
Big Data Real Time ArchitectureInputKafkaProcess, Analyze,VisualizeStorm/SparkStoreHDFS / HBaseCassandraDistributed scalable publish/subscribe system for BigDataA distributed platform fordoing analysis on stream ofmeasurement data in realtime Multi-tier Distributed Virtualized Parallel processing Mix workloads CloudData Lake – HDFS / HBASECassandra - Open Sourcedistributed DBMSwww.beznext.comBEZNext All Rights Reserved11
Performance Assurance During Application,Data and System Life Cycle AffectProve ofConceptsAutomationDevelopmentBig DataLife CycleCapacityPlanningBig Data ApplicationLife CycleWorkloadManagementTestingand BEZNext All Rights Reserved12
Value of Application PerformanceAssurance Optimization of design and development during application, data andsystems life cycle Optimization of performance management and workload management Optimization of Big Data infrastructure Set realistic expectations Enables verification Business process optimization Predictive and prescriptive analytics enables automatic proactiveperformance assurance process focusing on continuously meeting SLGs Reduce uncertainty and risk of performance surprises Enables collaborative capacity management process providing betteraliment between business and ITwww.beznext.comBEZNext All Rights Reserved13
Advanced AnalyticsDecision Optimization During Application and Data Lifecycle Descriptive analytics to identify significantchanges in applications performance,resource utilization and data usage profiles Diagnostic analytics to identify currentproblems and the root causes of thoseproblems Predictive analytics to answer What Ifquestions and to predict the outcome ofanticipated changes and identify potentialproblems Prescriptive analytics to evaluatedifferent options, provide proactiverecommendations and generate automatedadvice in order to set realistic expectations Control analytics compares the actualresults with expected in order to developcorrective actions and feed results into acontinuous management processwww.beznext.comBEZNext All Rights Reserved14
DATA COLLECTIONwww.beznext.comBEZNext All Rights Reserved15
Goals Organize continuous data collection from different systems Transform and aggregate data into workloads representing line ofbusiness with ability to drill down to users, applications, and so on For each workload, build performance, resource utilization and datausage profiles, and calibrate profiles to make data collected fromdifferent sources correspond to each otherwww.beznext.comBEZNext All Rights Reserved16
ApplicationsModificationVolume ofDatawww.beznext.comBudget for ITBEZNext All Rights ReservedCollaboration with IT inManaging ModificationRequirementsNew ApplicationsRequirementsNegotiation of SLGswith ITWorkloadAggregationData TransformationVerification againstMeasurement DataDefining OptionsSLG RequirementsGatheringInformationApplications andUsersIdentification of Lineof BusinessBusiness Data Collection StagesBusiness PlanPerformanceDataWarehouseNumber ofUsersData LakePerformanceRequirements17
PerformanceDataWarehouseData EDWDB2 EDWAgentManagerBig DataBig DataClustersClustersPerformanceEngineeringIT Data Collection StagesOther ITOther ITPlatformsPlatformswww.beznext.comBEZNext All Rights Reserved18
BEZNext Data Collection ComponentsAuto DiscoveryAgentWorkloadLinux AgentBig DataBig DataClustersClustersOther ITOther ITPlatformsPlatformsStorm AgentCassandraAgentWorkloadAggregationSpark AgentDataTransformationKafka AgentAgent ManagerTeradata,Teradata,Oracle,DB2Oracle,EDWDB2 EDWCharacterizationPerformanceDataWarehouseData LakeYARN ication &ControlTez AgentTeradata AgentOracle AgentOther Agentswww.beznext.comBEZNext All Rights Reserved19
Sources of Data Configuration Customer Ganglia / Ambari / Hadoop Hadoop and subsystems / YARN / Zookeeper Linux Resource Consumption Operating Systems Linux /proc directory (CPU, memory, IO and network traffic for each host as a whole andindividual processes) Windows, and so on HDFS NameNode (disk space) Performance Subsystems like YARN, Kafka, Spark, Storm, Cassandra, Hbase through API andJMX Teradata, Oracle, DB2, SQL Server Operating systemwww.beznext.comBEZNext All Rights Reserved20
BEZNext Agents Operating System Remote connection to the monitored server Data retrieval from existing 3rd party source Oracle JDBC connection to one of the instances associated with the monitored database.Retrieval of data from GV tables Teradata JDBC connection to the monitored Teradata system. Retrieval of data fromResusage, DBQL and TDWM Big Data Retrieval of metric sets from API or JMX interfaces to each specific technologyinstalled on the cluster (that is, YARN, Cassandra, Spark, Kafka, and so on)www.beznext.comBEZNext All Rights Reserved21
BEZNext Collectors / AgentsConfiguration - Overview Level - Detailed Activitywww.beznext.comBEZNext All Rights Reserved22
Sampling ExampleSample(s)SleepSample(s)SleepTimeDefault Properties for Oracle CollectionCollectorSleep IntervalProcess1 minNode5 minDevice (I/O)5 minInstance15 minSession2 minRequest10 secResponse Time15 minwww.beznext.comBEZNext All Rights Reserved23
BEZVision Parameterswww.beznext.comBEZNext All Rights Reserved24
Transformation / Profile CreationHourly Profiles Building Steps2PM . 3PMData retrieval for the 1PM to 2PMactivity takes place.Data is aggregated in hourly summaryviews.Workload assignment rules are appliedusing the default rule set and individualworkload profiles are created.Workload profiles are calibrated,the “automatic” profile is created and the1PM – 2PM data is made available in the product.www.beznext.comBEZNext All Rights Reserved25
Performance, Resource and DataUsage Profiles for Each WorkloadWorkloadAggregation RulesUsername ‘salesops’ andProgram ‘finapp.exe’PerformanceAverage Response TimeThroughputEtc .Resource UtilizationCPU UsageI/O RateMemory UsageNetwork Data Usage Read/WriteFrequency of DataAccessesParallelismJoin / Sorts OperationsEtc Workload Aggregation Rules are used toaggregate measurement data into workloadEach workload has performance, resourceutilization and data usage hourly profiles Line of Business (Marketing, Finance, and so on) Type of Activity Near Real Time, Batch, and so onwww.beznext.comBEZNext All Rights Reserved26
Level of Detail Depends on the problem you need to solve and availablesources of data Options: By System or Node By Subsystem By Workload / Application / Userwww.beznext.comBEZNext All Rights Reserved27
Frequency of Data Collection Frequency of sampling depends on data variability andoverhead of data collection Options Continuous data collection of basic performance and resourceconsumption data Variable rate of collection Collection of detail data only when anomaly is detected orpredictedwww.beznext.comBEZNext All Rights Reserved28
Data Transformation for Big Data Group individual data samples (like every minute) into modelingintervals (like every hour) Summarize resources consumed by child processes up to the parentprocess Map Linux processes to users, applications and Hadoop subsystems Match Linux processes with subsystems’ applications to create bothperformance and resource usage profiles Fill in information (“workload elements”) allowing grouping individualunits of work into business workloads Prepare configuration and workload information to import intoBEZVisionwww.beznext.comBEZNext All Rights Reserved29
What Data is Stored in Performance DataWarehouse? Aggregated data representing hourly workloads’ performance, resourceutilization and data usage profiles Results of auto-discovery characterizing configuration By streaming measurement data using Kafka, and by doing in-memorydata aggregation and calculation of hourly average, STD, 95 percentileand implementing diagnostic analysis with Storm or Spark, you canreduce overhead, implement near real time capacity management andreduce the volume of data storedwww.beznext.comBEZNext All Rights Reserved30
Service Level Goal (SLGs) Performance Response time, throughput Resource Utilization CPU, memory, SSD, HDD, network Data usage profile Read/write, parallelism, and so on Disk Space Usage Total, allocated, used Availability % of time when devices are available Reliability Frequency of errors and outages, including CPU, memory, SSD, HDD, network, software andapplications Power usage Correlation between power consumption and utilization of hardwarewww.beznext.comBEZNext All Rights Reserved31
Implementation for Big Data Shell scripts / Python scripts / C executables to collect Linux data oneach host Python scripts on the remote control node to collect the whole clusterand subsystem level data and to organize continuous data collectionprocess on changeable cluster configuration Java applications in scalable Kafka and Spark environment to pickdata from the cluster hosts and transform Additional module in BEZVision to import transformed data and createperformance and storage profileswww.beznext.comBEZNext All Rights Reserved32
Data Collection Summary BEZNext Agents incorporate Big Data capability toachieve scalable solutions supporting continuous 24 X 7data collection from distributed, multi-tier, parallelizedsystems, data transformation and processing BEZNext Agents incorporate advanced analytics to cleandata and reconstruct missing datawww.beznext.comBEZNext All Rights Reserved33
DataCollectionURLLinksLinux /proc directory: /html/proc.html YARN REST API: hadoop-yarn-site/WebServicesIntro.html Kafka JMX: ing Spark l Storm REST API: com/apache/storm/blob/master/STORM-UI-REST-API.md Cassandra ra/operations/ops monitoring c.html Hbase JMX: https://hbase.apache.org/metrics.html HDFS REST API: ct-dist/hadoop-hdfs/WebHDFS.htmlJMX: .3.0/bk hdfs admin tools/content/ch07.htmlwww.beznext.comBEZNext All Rights Reserved34
esource Utilization
Workload Aggregation andCharacterization Process Create Workload Aggregation rules Build Workload profiles– Performance, resource utilization and data usage profilesfor each workload Results of Workload Characterization are used for––––––Diagnostic and root cause analysisDetermining seasonal peaks and workload forecastingWorkload managementPerformance ManagementCapacity planningGenerating prescriptionswww.beznext.com BEZNext All Rights Reserved36
What is a Workload? A workload represents aggregated activity of a group of users orapplications supporting a specific line of business, business function ordepartment Workloads characterization provides an integrated view of the businessdemand for IT Resources and Data on one hand and level of service orperformance provided by IT in servicing the workload Each Workload has unique performance, resource utilization and datausage profiles– Performance profile – the average response time and throughput– Resource utilization profile – average CPU utilization, I/O rate, Memory anddisk utilization, level of concurrency, level of parallelism and networkutilization– The data usage profile includes the frequency and type of data access Increase in number of users, volume of data, implementation of newapplications and modification of existing applications changes workloads’profileswww.beznext.com BEZNext All Rights Reserved37
PerformanceDataWarehouseData radata,Oracle,DB2Oracle,EDWDB2 EDWAgent ManagerBig DataBig DataClustersClustersPerformanceEngineeringWorkload Characterization ProcessOther ITOther ITPlatformsPlatformsAggregate Data into WorkloadsEach workload represents a line of business, businessprocess, department or group of usersMetrixPerformance ProfileResponse timeThroughputResource Utilization ProfileCPUI/OMemoryInternode communicationTotal CPU Seconds ConsumedTotal I/O OperationsTotal number of Requests - Throughput (requests/second)Data Access ProfileParallel Sessions (concurrent connections)Read/WriteLevel of parallelismDelay Time (seconds)www.beznext.com BEZNext All Rights Reserved38
Workload Aggregation RulesHourly Profiles Input for workload aggregation: transformed measurement data andWorkload Aggregation Rules (WAG) WAG use:– Users’ names, application/program names or other common parameters– Cluster analysis results based on performance and usage of resources WAG aggregate detail measurement data into business workloadswww.beznext.com BEZNext All Rights Reserved39
Workload Aggregation (WAG)www.beznext.com BEZNext All Rights Reserved40
Workload Aggregation Challenges End effect Distribute delta between OS andsubsystem measurement databetween workloads or create aseparate workload for OS ownactivity, or unrecognized activity(misc workloads in BV) Coordination of workloadsbetween tiers, clustersOSwww.beznext.com BEZNext All Rights Reserved41
Performance, Resource and Data UsageProfiles for Each WorkloadWorkloadAggregation RulesUsername ‘salesops’ andProgram ‘finapp.exe’PerformanceAverage Response TimeThroughputResource UtilizationCPU UsageI/O RateMemory UsageNetworkData UsageRead/WriteFrequency of Data AccessesParallelismJoin / Sorts Operations Workload aggregation rules are used toaggregate measurement data into WorkloadEach workload has performance, resourceutilization and data usage hourly profiles Line of business (Marketing, Finance, etc.)Type of activity (near real time, batch, etc.)www.beznext.com BEZNext All Rights Reserved42
Output of the WorkloadCharacterization Workload characterization is performed continuously 24 X 7Performance, resource utilization and data usage profiles are created hourly foreach workload.Performance profile includes workload average response time, user think time andthroughput during different representative time intervals; for example, prime shiftduring holiday season, prime shift end of month processing and typical week dayWorkload’s resource usage profile of each workload includes average number ofactive users, average priority of requests within the workload, average CPUutilization by application, inter-node utilization, I/O rate to disks, read/write ratio,average level of parallelismWorkload’s sata usage profile includes the list of files, databases and tablesaccessed by applicationsDisk space usage is determined periodicallyAdvanced analytics identify the trends and significant changes in performance,usage of resources and data, enable root cause analysis and focus performancetuning on the most critical problems affecting performance of the most criticalworkloadswww.beznext.com BEZNext All Rights Reserved43
Demo Examples of workload characterization results Examples of rules for data aggregationwww.beznext.com BEZNext All Rights Reserved44
Workloads Profileswww.beznext.com BEZNext All Rights Reserved45
Scatter Plot Matrix Grouped /binned into hexagonal tiles46BorisBEZNextZibitsker,All RightsReservedwww.beznext.comAll RightsReserved46
CPU Utilization by Business WorkloadsETL Sales and ETL Marketing Use Almost 40% of Resourceswww.beznext.com BEZNext All Rights Reserved47
Boris Zibitsker, All rights ReservedDetermining Anomalies Statistical Process Controlwww.beznext.comAll Rights ReservedBEZNext All Rights Reserved48
Root Cause AnalysisRandom ForestIn this example, using aRandom Forest modelyielded a model with similarfits, but different insightsinto the data.We can see which variablesthe model found to beimportant.Random Forest models usean ensemble of trees tomake predictions.49www.beznext.comBEZNext All Rights Reserved
Root Cause Analysis – Decision TreeLeaf page and branches identify the causewww.beznext.com BEZNext All Rights Reserved50
Workload Characterization Data Aggregation Building workloads’ profiles Performance Resource utilization Data usage Results are used as input for: Workload forecasting Performance management Workload management Capacity planningwww.beznext.comBEZNext All Rights Reserved51
PERFORMANCE MANAGEMENT
Performance Management Descriptive Analytics Current and past performance Diagnostic Analytics Anomalies detection Root cause Analysis Predictive Analytics Discover future bottlenecksCurrent and Past PerformanceDiagnostic analytics identifies significant changes inperformance and resource utilization of the individualworkloadswww.beznext.comBEZNext All Rights Reserved53
Performance Analysiswww.beznext.comBEZNext All Rights Reserved54
Anomaly Detection Diagnostic analytics identifiesanomalies Determining significantChanges with RT, throughputand resource utilizationdiagnostic analyticswww.beznext.comBEZNext All Rights Reserved55
Root Cause Analysis Determine causes ofperformance degradation Decision trees and Logistic regression analysis Predict future bottlenecks Predictive AnalyticsDecision Tree - Leaf page and branches identify the rootcausewww.beznext.comBEZNext All Rights Reserved56
PriorityResource AllocationWORKLOAD MANAGEMENTConcurrency
Example of Workload Managementby QueuesByOrganization(100%) By Organization 100% Marketing 33% Finance 33% Sales 33%Marketing(33%)Finance (33%) By Type of Workload 100% Near Real Time 70% Batch 30%Sales (33%)By Type ofWorkload (100%)Near Real Time(70%)Batch (30%) Hybrid 100% Marketing 20%Hybrid (100%) Batch 15% Real-Time 5% Finance(40%)Marketing (20%)Finance – 40%Sales (Batch40%) Real-Time 10% Batch 30% Batch (15%)Real Time (5%)Batch (30%)Real Time (10%)Sales – Batch 40%www.beznext.comBEZNext All Rights Reserved58
Resource Manager Schedulers FIFO Scheduler Processing jobs in order Capacity Scheduler (Default) Queue shares as percentage of clusters FIFO scheduling within each queue Supporting preemption Fair Scheduler Fair to all userswww.beznext.comBEZNext All Rights Reserved59
YARN Capacity SchedulerMarketingFinanceSalesSet limits on capacity: Minimum capacity for the queueQueue 3Queue 2Queue 150%resources) for a queue Resource elasticity when not beingused by other queues Minimum user limits – user sharingGuaranteed Resources30% Maximum capacity (% of clusterfor a given queue20% User limit factor – maximum queuecapacity that one user can take up Application limit – maximum # ofapplications submitted to one queuewww.beznext.comBEZNext All Rights Reserved60
Example of Predicting WorkloadConcurrency Change Impactwww.beznext.comBEZNext All Rights Reserved61
Example of Predicting WorkloadPriority Change Impactwww.beznext.comBEZNext All Rights Reserved62
CAPACITY PLANNING
ceResource UtilizationData UsageConsolidatedHardwarePerformanceNumber of NodesType of NodesWarehouse edictedResponse TimePerformance Prediction andPrescriptionSoftwareLinuxYARN SettingsKafka, Spark, Storm,Cassandra, etcWorkloadForecastingWorkload GrowthVolume of mber of UsersVolume of DataBEZNext All Rights ReservedPrescriptions64
Collecting Data and Modeling inTest plicationsand DataTesting andModelingDataWarehousesand DataMartswww.beznext.comBEZNext All Rights Reserved65
Predicting How New Applications willPerform in Production EnvironmentNew ApplicationsExisting ApplicationsModelingwww.beznext.comBEZNext All Rights ReservedDataCollection66
Dynamic Capacity ManagementPrescriptionPredictionYARN Settings:WorkloadsSubsystemQUEUER&DYARNQueue R&DMKTYARNQueue MKTSalesYARNQueue SLSFinanceYARNQueue FINHRYARNQueue HR Scheduler Type Queue Structure Priority ResourceLimits Concurrency Container settingwww.beznext.comCluster NodesClusterContainerNodes2Container 1YARN SchedulerKafkaSparkContainer 3SparkContainer nCassandraBEZNext All Rights Reserved67
Capacity PlanningPredict the impact of workloadand volume of data growthDetermine when workloads SLGswill not be met Long Term Planning Apply Predictive Analytics todetermine number of Nodes inCluster required to support expectedworkload and volume of data growth Predict how new application willperform on production system Dynamic – Real Time CapacityPredict how newapplication will affectperformance ofexisting applicationsPlanning Apply Prescriptive Analytics toevaluate options and determine howto dynamically change YARNSettings, including Containers,Queues and Scheduler to meetindividual workloads SLGsPredict the impact ofproposed changes Set realistic expectations Verify resultswww.beznext.comBEZNext All Rights Reserved68
Predicting New ApplicationImplementation ImpactProductionTestNew GB Data ingModeling Testand ProductionSystemsPredicting erificationYARN Settings:NewCurrent Sales, Mkt, HR, etcApplicationMore DataMore UsersLong TermCapacity PlanningDynamic Containers Queues SchedulerCapacity Planningwww.beznext.comPBBEZNext All Rights ReservedYARN Settings: Containers Queues Scheduler69
VERIFICATION ANDAUTOMATION
Verification – Actual vs Expected(A2E)www.beznext.comBEZNext All Rights Reserved71
SUMMARY
Value of Application PerformanceAssurance Optimization of Performance Management, Workload Managementand Capacity Planning Decisions during Application, Data andSystems Life Cycle Set Realistic Expectations Enables Verification Automation Predictive and Prescriptive analytics enables automatic proactive PerformanceAssurance process focusing on continuous meeting SLGs Reduce uncertainty and risk of performance surprises Collaboration Better aliment between business and ITwww.beznext.comall rights reserved
References B. Zibitsker, Big Data Advanced Analytics, Minsk 2016, Key Note Presentation on Big Data AnalyticsB. Zibitsker, CMG 2016, Performance Assurance for Real Time ApplicationsB. Zibitsker, CMG 2016, Enterprise Performance Assurance PlatformB Zibitsker, IEEE Conference, Delth Netherlands, March 2016, Big Data Performance AssuranceB. Zibitsker, Big Data Predictive Analytics Conference, Minsk 2015, Key note presentation “Role of Big DataPredictive Analytics”B. Zibitsker, Big Data Predictive Analytics Conference, Minsk 2015, Workshop on “Big Data Predictive Analytics”B. Zibitsker, Big Data Conference, Riga 2014, “Application of Predictive Analytics for Better Alignment of Businessand IT”B. Zibitsker, T. Jung, Teradata Partners, 2012, “Collaborative Capacity Management”B. Zibitsker, A. Lupersolsky, OOW 2009, “Modeling and Optimization in Virtualized Multi-tier Distributed Environment”B. Zibitsker, Teradata Partners 2008, “Proactive Performance Management of Data Warehouses with MixedWorkloads”B. Zibitsker, DAMA 2007, “Enterprise Data Management and Optimization”J. Buzen, B. Zibitsker, CMG 2006, “Challenges of Performance Prediction in Multi-tier Parallel ProcessingEnvironments”B. Zibitsker, CMG 2008, 2009 “Hands on Workshop on Performance Prediction for Virtualized Multi-tier DistributedEnvironments”www.beznext.comBEZNext All Rights Reserved74
Boris Zibitsker, PhD Founder and CEO of BEZNext, 2011 - presentCurrent focus of research is on applying predictive and prescriptive analytics for optimization of business and IT decisionsduring applications and data life cycleManage development of the Performance Assurance technology incorporating advanced analytics for optimization of Big Dataand Data Warehouse applications in complex multi-tier, distributed, virtualized, parallel processing environmentConsulted many of Fortune 500 companiesCTO of Modeling and Optimization at Compuware (2010-2014)Participated in development of Application Performance Management software incorporating Machine Learning algorithms forperformance and availability problems detection, and root cause analysis determination for web applicationsFounder, President and Chairman of BEZ Systems (1983 - 2010), acquired by Compuware in 2010Managed development of BEZVision Performance Prediction and Capacity Management software for Teradata, Oracle, DB2and SQL ServersPerformance Analyst:Started out as engineer at Computer Systems Research Institute working on modeling and performance evaluation of largecomputer systems and applying modeling results for optimization of jobs scheduling and storage performance managementWorked in capacity management departments at FNBC and CNA Insurance company in ChicagoAdjunct Associate Professor, DePaul University in Chicago (1983 – 1990)Taught graduate courses on Modeling of Computer Systems, Queueing Theory with Computer Applications, ComputerCommunication Systems Design and AnalysisTaught seminars at Northwestern University, University of Chicago and Relational Institute - North and South America,Europe, Asia, and AfricaAuthor of papers on applying modeling and optimization for performance evaluation, performance assurance, performancemanagement, workload management and capacity planning for Big Data and Data Warehouse environmentsEducation: MS and PhD research at BSUIR and NIIEVMwww.beznext.comBEZNext All Rights Reserved75
ARE THERE ANY QUESTIONS?
Problem from the Business Prospective Business needs To make effective business decisions fast To increase profitability and reduce IT cost Business requirements to IT Activity of group of business users, customers and vendors using applications of the line of business is a workload Applications should: Provide Information necessary to support line of business decisions
