A Multithreading C# Data Synchronization Program and Its Realization Course: ECE 1747H Parallel Programming Professor: Christiana Amza Student / Presenter:

A Multithreading C# A Multithreading C# Data Synchronization Program Data Synchronization Program

and Its Realizationand Its Realization

Course: ECE 1747H Parallel ProgrammingCourse: ECE 1747H Parallel ProgrammingProfessor: Christiana AmzaProfessor: Christiana AmzaStudent / Presenter: Bin LiStudent / Presenter: Bin Li

Dec. 12, 2006 @ University of TorontoDec. 12, 2006 @ University of Toronto

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto2

AgendaAgenda

BackgroundBackground

Problem & SolutionProblem & Solution

Parallel ImplementationParallel Implementation

Performance MeasuringPerformance Measuring

Other ApproachesOther Approaches

Future WorkFuture Work

Q & AQ & A

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto3

Background Background (Company & Project)(Company & Project)

““Retail Value Canada Inc.”Retail Value Canada Inc.”Markham-based Specialty RetailerMarkham-based Specialty Retailer

384 stores in Canada & USA, 30K types of 384 stores in Canada & USA, 30K types of itemsitems

Head Office-side information maintained in Head Office-side information maintained in WindowsWindows

Store-side information maintained in UnixStore-side information maintained in Unix

Data synchronization is neededData synchronization is needed

Data type: product code, status, cost, price, Data type: product code, status, cost, price, promo, deal, subsidy, vendor, warehouse, promo, deal, subsidy, vendor, warehouse, etc. (by item, by store)etc. (by item, by store)

Current application: Current application: iSynciSync (developed in (developed in 2000 in Visual C# 1.0)2000 in Visual C# 1.0)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto4

Background Background (System Architecture)(System Architecture)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto5

AgendaAgenda

BackgroundBackground Problem & SolutionProblem & Solution

Parallel ImplementationParallel Implementation

Performance MeasuringPerformance Measuring

Other ApproachesOther Approaches

Future WorkFuture Work

Q & AQ & A

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto6

Problem & SolutionProblem & SolutionScheduled Data Synchronization (Scheduled Data Synchronization (iSynciSync process) starts at 10 pm, and ends at process) starts at 10 pm, and ends at 12 am12 am

iSynciSync extracts and transforms data from extracts and transforms data from Windows into ASCII file (.dat), and sends Windows into ASCII file (.dat), and sends it to Unixit to Unix

Mass data modification takes Mass data modification takes iSynciSync quite a long time (4-5 hours) to run, quite a long time (4-5 hours) to run, which is over 2-hour schedule limit which is over 2-hour schedule limit

The latest change (i.e. prices) in head The latest change (i.e. prices) in head office cannot reach stores before the office cannot reach stores before the opening hour of the next business dayopening hour of the next business day

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto7

Problem & Solution Problem & Solution (cont’d)(cont’d)

The store-side information delay causes The store-side information delay causes inaccurate sales information in retail inaccurate sales information in retail storesstores

Bottleneck: Bottleneck: iSync iSync (only 10% CPU usage (only 10% CPU usage on a 4-CPU database server)on a 4-CPU database server)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto8

Problem & Solution Problem & Solution (cont’d)(cont’d)

Sequential programSequential programiSynciSync generates .dat file generates .dat file by each storeby each store, , which is slowwhich is slow

Parallel solutionParallel solutionImplementing Implementing qSyncqSync to replace to replace iSynciSync (using (using Microsoft C# multithreading)Microsoft C# multithreading)

Parallelly generating .dat file Parallelly generating .dat file by store by store groupsgroups

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto9

AgendaAgenda

BackgroundBackground Problem & SolutionProblem & Solution Parallel ImplementationParallel Implementation

Performance MeasuringPerformance Measuring

Other ApproachesOther Approaches

Future WorkFuture Work

Q & AQ & A

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto10

Parallel ImplementationParallel ImplementationDevelopment EnvironmentDevelopment Environment

Design/UML Tool: Microsoft Visio 2003Design/UML Tool: Microsoft Visio 2003

Development Tool: Microsoft Visual Development Tool: Microsoft Visual Studio .NET 2003Studio .NET 2003Programming Language: Visual C# 2.0 Programming Language: Visual C# 2.0 ((multithreading similar to Linux PThreads)multithreading similar to Linux PThreads)

Parallelization StepsParallelization StepsStore Data SegmentationStore Data Segmentation

Parallel Data ProcessingParallel Data Processing

Result Data Consolidation Result Data Consolidation

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto11

Parallel Implementation Parallel Implementation (cont’d)(cont’d)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto12

User Interface User Interface (Screen 1)(Screen 1)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto13

User Interface User Interface (Screen 2)(Screen 2)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto14

Sample CodeSample Code

using System.Threading;using System.Threading;

private int getNbrOfInstance()private int getNbrOfInstance(){{ //...//...

string sqlStmt = "select cast(RBSValue as int) from RulesBasedSystem " +string sqlStmt = "select cast(RBSValue as int) from RulesBasedSystem " + "where RBSTxt = 'HISSPNbrOfInst' and RBSScopeKey = 'Retail Value'";"where RBSTxt = 'HISSPNbrOfInst' and RBSScopeKey = 'Retail Value'";

//...//...}}

HISSPCLPSyncComponent clpComponent = null;HISSPCLPSyncComponent clpComponent = null;clpComponent = new HISSPCLPSyncComponent();clpComponent = new HISSPCLPSyncComponent();

ThreadStart threadDelegate=null;ThreadStart threadDelegate=null;Thread threadObj=null;Thread threadObj=null;

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto15

Sample Code Sample Code (cont’d)(cont’d)

for (int i=0; i<dtb.Rows.Count; i++)for (int i=0; i<dtb.Rows.Count; i++){{ //...//...

threadDelegate = new ThreadStart(clpComponent.ExtCLPPrice);threadDelegate = new ThreadStart(clpComponent.ExtCLPPrice);threadObj = new Thread(threadDelegate);threadObj = new Thread(threadDelegate);threadObj.Name = Convert.ToString(i);threadObj.Name = Convert.ToString(i);threadList.Add(threadObj);threadList.Add(threadObj);//Start the thread//Start the threadthreadObj.Start();threadObj.Start();

}}// Join the threads// Join the threadsfor (int i = 0; i<dtb.Rows.Count; i++)for (int i = 0; i<dtb.Rows.Count; i++){{

threadObj = (Thread) threadList[i];threadObj = (Thread) threadList[i];threadObj.Join();threadObj.Join();

}}while(j>0) //Approach #3while(j>0) //Approach #3{{

lock(this)lock(this){{ consolidateCLPPrice(baseStoreId[j], itemBaseId[j], marketZoneId[j], itemPackId[j]);consolidateCLPPrice(baseStoreId[j], itemBaseId[j], marketZoneId[j], itemPackId[j]);}}j--;j--;

}}

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto16

AgendaAgenda

BackgroundBackground Problem & SolutionProblem & Solution Parallel ImplementationParallel Implementation Performance MeasuringPerformance Measuring

Other ApproachesOther Approaches

Future WorkFuture Work

Q & AQ & A

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto17

Performance MeasuringPerformance MeasuringTesting EnvironmentTesting Environment

Database ServerDatabase ServerIntel Xeon CPU 2.40 GHz, Intel Xeon CPU 2.40 GHz, 44 CPUs, 3GB RAM CPUs, 3GB RAM

Windows 2000 w/SP4, MS SQL Server 2000Windows 2000 w/SP4, MS SQL Server 2000

Subset of real production dataSubset of real production data

Web/Application ServerWeb/Application ServerIntel Pentium4, Intel Pentium4, 22 CPU 3.40Ghz ( CPU 3.40Ghz (HTHT), 2GB RAM), 2GB RAM

Windows XP w/SP2, IIS 5.1Windows XP w/SP2, IIS 5.1

Performance CountersPerformance CountersCPU % UsageCPU % Usage

Execution TimeExecution Time

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto18

Performance ComparisonPerformance ComparisonNumber of ThreadsNumber of Threads CPU Usage / Execution Time (sec)CPU Usage / Execution Time (sec)

1 (sequential)1 (sequential) 10%10%

2552255222 25%25%

15451545 55 75%75%

5275271010 93%93%

4614612020 100%100%

370370

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto19

Performance Comparison Performance Comparison (cont’d)(cont’d)

CPU % UsageCPU % Usage

Execution TimeExecution Time

0

20

40

60

80

100

1 2 3 5 10 15 20

Number of Threads

Parallel Data Synchronization(CPU % Usage Comparison)

CPU Usage (%)

0

5001000

1500

20002500

3000

1 2 3 5 10 15 20

Number of Threads

Parallel Data Synchronization(Execution Time - seconds)

Execution Time(seconds)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto20

AgendaAgenda

BackgroundBackground Problem & SolutionProblem & Solution Parallel ImplementationParallel Implementation Performance MeasuringPerformance Measuring Other ApproachesOther Approaches

Future WorkFuture Work

Q & AQ & A

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto21

Other Approaches Other Approaches (Approach #2)(Approach #2)

““Locking Temp Files”Locking Temp Files”

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto22

Other Approaches Other Approaches (Approach #2 cont’d)(Approach #2 cont’d)

““Locking Temp Files”Locking Temp Files”All threads write to single .dat fileAll threads write to single .dat file

Using lock for file appendingUsing lock for file appending

Result: bad as sequentialResult: bad as sequential

Explanation: same disk file cannot be Explanation: same disk file cannot be shared simultaneously by different threads, shared simultaneously by different threads, needs to close/re-open (different from needs to close/re-open (different from shared memory)shared memory)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto23

Other Approaches Other Approaches (Approach #3)(Approach #3)

““Locking Temp Tables”Locking Temp Tables”

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto24

Other Approaches Other Approaches (Approach #3 cont’d)(Approach #3 cont’d)

““Locking Temp Tables”Locking Temp Tables”All threads share single temporary All threads share single temporary database tabledatabase table

Using lock for table record insertingUsing lock for table record inserting

Result: much better than sequential, not as Result: much better than sequential, not as good as the Main Approachgood as the Main Approach

Explanation: database server has enough Explanation: database server has enough memory; lock brings slight delaymemory; lock brings slight delay

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto25

AgendaAgenda

BackgroundBackground Problem & SolutionProblem & Solution Parallel ImplementationParallel Implementation Performance MeasuringPerformance Measuring Other ApproachesOther Approaches Future WorkFuture Work

Q & AQ & A

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto26

Further WorkFurther Work

Database ParallelismDatabase ParallelismUpgrading SQL Server 2000 to 2005Upgrading SQL Server 2000 to 2005

Migrating C# code of data synchronization Migrating C# code of data synchronization to database stored procedures, optimizing to database stored procedures, optimizing SQL queriesSQL queries

Changing temporary table(s) to permanent Changing temporary table(s) to permanent schemaschema

Using SQL Server Integration Services (SSIS Using SQL Server Integration Services (SSIS 2005) to do parallel data load & 2005) to do parallel data load & transformationtransformation

Accessing permanent table (which contains Accessing permanent table (which contains final data to be synchronized) to final data to be synchronized) to generate .dat filegenerate .dat file

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto27

Q & AQ & A

Thanks!Thanks!

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto28

Additional Slide for Q&A Additional Slide for Q&A (C#)(C#)

C# (pronounced “C Sharp”)C# (pronounced “C Sharp”)Microsoft .NET Framework-compliant Microsoft .NET Framework-compliant languagelanguage

Simple, modern, object oriented Simple, modern, object oriented programming language derived from C and programming language derived from C and C++C++

Aims to combine the high productivity of Aims to combine the high productivity of Visual Basic and the raw power of C++.Visual Basic and the raw power of C++.

C# vs JavaC# vs JavaSimilar but not same in language Similar but not same in language specificationsspecifications

Compilation: C# to Microsoft Intermediate Compilation: C# to Microsoft Intermediate Language (MSIL), and Java to Java bytecode Language (MSIL), and Java to Java bytecode

Running: C# in Common Language Running: C# in Common Language Runtime (CLR), Java in Java Virtual Machine Runtime (CLR), Java in Java Virtual Machine (JVM)(JVM)

Parallel Programming Professor: Christiana Amza Student: Bin Li Dec.12, 2006 @ University of Toronto29

Additional Slide Additional Slide for Q&A for Q&A (Main Approach vs Approach #3)(Main Approach vs Approach #3)

0

200

400

600

800

Number of Threads

10 20

Parallel Data Synchronization(Execution Time - Seconds)

Main Approach (no datalock)

Approach #3 (lockingtemp table)