Export (0) Print
Expand All
2 out of 2 rated this helpful - Rate this topic

Appendix - Scalability, Performance, and Capacity Testing

On This Page

Introduction Introduction
Testing Methodology Testing Methodology
Test Environment Test Environment
Test Configuration Summary Test Configuration Summary
Test Plan Test Plan
Hardware and Software Details for Tests Hardware and Software Details for Tests
Test Results and Analysis Test Results and Analysis
Key Findings Key Findings
Recommendations for Capacity Planning Recommendations for Capacity Planning
Conclusion Conclusion

Introduction

Microsoft and Infosys, in partnership with a leading automotive manufacturer in Europe and Intel Solution Services, have developed a set of comprehensive tests aimed at validating the scalability of the Microsoft operating platform on Intel architecture to support the CATIA V5 application.

Extensive tests with various combinations of components and scenarios were executed at the Intel Solution Center in Munich, Germany with the objectives of validating the feasibility of Code Serving the CATIA V5 application and identifying an optimal architecture for deploying CATIA V5 using the Windows operating platform on Intel architecture. A dedicated lab environment with isolated infrastructure ensured that the tests were not influenced by external parameters. A worst-case scenario methodology was adopted for the tests to simulate the maximum simultaneous load, utilizing multiple workstations accessing the server at the same time, supplemented with automated macros ensuring concurrency. The test results were analyzed and are documented in detail in the subsequent sections of this chapter.

The results of these exhaustive tests bring to light some key findings:

  • CATIA V5 in Windows on the latest 32-bit Intel architecture is an extremely viable, scalable, and cost-effective alternative to the existing UNIX environment.

  • Code serving of CATIA V5 utilizing Windows 2000 Advanced Server is comparable in performance to locally installed CATIA V5 code on the workstations, and it reduces management and maintenance headaches significantly.

  • The use of Offline Folders provides efficient caching of the code on individual workstations, thus reducing the CPU, disk, memory, and network load on the servers distributing the binaries.

  • The tests also proved that the use of Distributed File Systems (DFS) further enhances the scalability of the servers while providing mission-critical high availability of these servers to ensure continued and uninterrupted access to the data.

  • In addition, the use of Code Serving allows simple switching between versions of CATIA with minimal user disruption.

In summary, network connectivity between the Code Servers and the workstations is the limiting factor for Code Serving of CATIA V5 on a Windows operating environment and supporting infrastructure. However, the network limitation can be alleviated through the use of the features mentioned above to arrive at an architecture capable of providing comparable performance to local installation of CATIA while providing higher levels of availability.

The results from the proof of concept tests at the Intel Solution Center have gone a long way in confirming the industry-wide belief that CATIA deployment on a Windows operating environment based on an Intel architecture is highly reliable and scalable, reduces the total cost of ownership (TCO), and is the optimal architecture for the future, thereby paving the way for its adoption and implementation in industry.

Testing Methodology

The subheadings below list the methods used in the test.

Objective

The tests provided the first-ever extensive scalability and sizing information for the code-serving functionality of Dassault Systems CATIA V5 software on the Microsoft operating platform based on Intel architecture.

The objectives of the tests were to evaluate the following:

  • The scalability of a CATIA V5 Code Server. Dassault Systems has provided a Code Serving capability to CATIA V5 since Release 9. The tests were designed to establish the feasibility of the Code Serving function and to define the number of workstations a Code Server could support in a real life scenario. Moreover, recommendations for infrastructure sizing needed definition.

  • Impact of Offline Folders. Windows 2000 Server and Windows XP provide the Offline Folders feature. Here, a client caches any files it uses from the server, on the local hard disk. This allows the file to be opened more quickly and with little network traffic on subsequent occasions. The feasibility and impact of effectively utilizing this feature in a CATIA code-serving environment were to be validated.

  • DFS. Windows 2000 Server provides the DFS feature to access multiple Code Servers. CATIA code was installed on multiple servers and a common share was published using DFS. The DFS feature, designed to automatically guide workstations to pick CATIA code from the next available server at any given time, was to be verified.

The server and client performance needed to be compared to a locally installed CATIA system accessing local CATIA data on a standalone workstation. These scenarios were tested at network speeds of 10 Mbps and 100 Mbps.

Configuration of Macros and Scripts

Macros and CATIA scripts were used to simulate the real-world user behavior of CATIA V5 R9 in several different scenarios. The macros were developed using CATIAs CATScripting tools, Windows Shell scripts, and batch command files. Windows tools were used to automate the tests and log the data for analysis.

Cases were planned to test performance in scenarios like Code Serving, Offline Folders and DFS compatibility with CATIA, as well as the performance of the workstations, Code Servers, and data servers. Macros and scripts executed the tests and recorded the events, times, and other key parameters during the test.

Configuring CATIA for Tests

License User Management (LUM) Configuration

Before CATIA can be accessed for the first time, a license server configuration is needed on the server and the workstation. For more information about CATIA license management, see Chapter 6, "CATIA Data Migration." The concurrent CATIA licenses were installed on a LUM server. Workstations were configured with the license server information on the LUM client software to allow them to pick the license when CATIA starts up.

CATIA Code Server Tests

CATIA V5 R9 SP1 Product AL2 was installed on the Code Servers. For a detailed discussion of this style of setup see Chapter 4, "CATIA Code Server Installation." The code files were located in the B09_SP1 folder and the environment file under CATENV\B09_SP1. The code folder B09_SP1 and the environment folder CATENV\B09_SP1 were shared to enable all client machines to access CATIA from the server. In the case of tests with client side caching, Offline Folders were enabled with an Active Directory group policy, and the caching option was turned on for the folder B09_SP1 with automatic caching for the programs option. The CAD data (CATPart and CATDrawing) was located under a specific share on the data server.

CATIA Local Tests

To compare the above setup with a local installation, CATIA V5 R9 SP1 Product AL2 was installed on the local drive of a workstation. The code files were located under the B09_SP1 folder and the environment file in the location CATENV\B09_SP1. The CAD data (CATPart and CATDrawing) were located on the local drive of the workstation.

CATIA with DFS Tests

A DFS setup is similar to a Code Server setup. CATIA code folders were located on multiple machines and a common DFS link was available for the workstations to access CATIA.

Test Environment

The test setup was hosted in the Intel Solution Center in Munich, Germany. The tests used 24 client workstations and 6 servers. The servers were Windows 2000 Advanced Servers SP3 and the clients were running Windows 2000 Professional Edition SP3.

A Windows 2000 domain, with DNS name ProtoV5.fr, was set up for all the components in the infrastructure. Server IP addresses were static, whereas IP addresses for the workstations were configured to be automatically allocated using a DHCP server. Additionally, two servers were used for CATIA Code Serving and one more was used for housing the CAD data. A (LUM) and two Active Directory servers were also part of the setup. The servers were connected through an Intel Netstructure 480T Routing Switch and the workstations through an Intel Express 510T Switch. Both switches were interlinked using a Gigabit Fiber connection.

Further details about the hardware, software, and the testing tools used in the project are documented in subsequent sections.

Network Schematic Diagram

Figure A.1 shows the network infrastructure of the testing environment in the lab:

Figure A: .1 Network Schematic

Figure A: .1 Network Schematic

Note: Thick lines in the figure indicate 1Gbps links, while thin lines indicate 100 Mbps links.

Hardware and Software Components

The hardware and software components used in the scalability test are listed below.

Hardware Components

Servers Supermicro Intel Xeon DP

  • Data Server: Stores CATIA Models and associated data accessed by all clients.

  • License Server: Issues CATIA licenses to the clients to run CATIA.

  • Active Directory/DNS: Acts as Domain Controller and resolves host names to IP addresses.

  • DHCP Server: Issues IP addresses to the clients.

Code Server Intel Xeon MP

Central installation of CATIA; shares the CATIA-code with clients.

Workstations HP Workstation x2100

24 workstations used as clients for accessing CATIA.

Network Components

The network connectivity between the Servers and the workstations was established through Intel switches. The Data Server and Code Servers were connected with 1 Gbps links and the other connections were 100 Mbps. A list of the network switches used in the testing environment follows:

Switches

  • Intel Express 510T Switch

  • Intel Netstructure 480T Routing Switch

Software Components

  • Windows 2000 Advanced Server SP3

  • Windows 2000 Professional Edition SP3

  • CATIA V5 R9 SP1

  • LUM 4.6.4

Testing Tools

Testing the scalability feature of CATIA for Code Serving, Offline Folders, and DFS was carried out entirely using macros and batch commands based on the Windows platform. No specific commercially available tools were used for the testing exercise. Real world scenarios were simulated through the use of user-defined scripts and CATIA macros in addition to certain Windows Resource Kit utilities.

CATIA macros were used to automate the work of the CATIA application. Windows scripts were created to modify the registry settings as required. Test metrics were defined and selected performance counters were captured automatically during the test runs. Multiple iterations and other sequential operations for the tests were included in batch files and script files to avoid manual intervention, thus ensuring the accuracy of the results.

Tests were scheduled with the Windows Scheduler, and performance counters were captured using the Windows command line utility TypePerf.exe. Registry values were modified in the scripts to run automatically for a pre-defined number of iterations (10).

Test Configuration Summary

CATIA performance was analyzed under many different scenarios. Test combinations included location of the code, network speed, use of Offline Folders, number of Code Servers (DFS), number of CPUs on the Code Servers, amount of memory on the workstations, and number of clients simultaneously accessing CATIA.

The table below presents detailed information for each test. The number of iterations was 10 for all tests. Key statistics, such as CATIA start time, Change Work Bench, and Drawing time, were used for the analysis of the results.

Table A.1 Test Scenarios

Test No

Test Scenario

Test 01

Local Data; Local Code SP1 (2 Workstations 512 MB RAM, 1 Workstation 1 GB RAM)

Test 02

Local Data; Local Code SP4; 1 Workstation

Test 03

Data Server; Code Server SP1; 100 Mbps; 1 Workstation

Test 04

Data Server; Code Server SP1; 100 Mbps; 1 Workstation; Offline Folders

Test 05

Data Server; Code Server SP1; 100 Mbps; 10 Workstations

Test 06

Data Server; Code Server SP1; 100 Mbps; 10 Workstations; Offline Folders

Test 07

Data Server; Code Server SP1; 100 Mbps; 24 Workstations

Test 08

Data Server; Code Server SP1; 100 Mbps; 24 Workstations; Offline Folders

Test 09

Data Server; 2 Code Servers SP1; 100 Mbps; 10 Workstations; DFS

Test 10

Data Server; 2 Code Servers SP1; 100 Mbps; 24 Workstations; DFS

Test 11

Data Server; Code Server SP1; 10 Mbps; 10 Workstations

Test 12

Data Server; Code Server SP1; 10 Mbps; 10 Workstations; Offline Folders

Test 13

Data Server; Code Server SP1; 10 Mbps; 24 Workstations

Test 14

Data Server; Code Server SP1; 10 Mbps; 24 Workstations; Offline Folders

Test 15

Data Server, Code Server SP1; 100 Mbps; 1 Workstation; 256 MB RAM on the workstation

Test 16

Data Server; Code Server SP1; 100 Mbps; 24 Workstations; 2 CPUs on the Code Server

Test 17

Data Server; Code Server SP1; 100 Mbps; 24 Workstations; 4 CPUs on the Code Server

Test 18

Data Server; Code Server SP1; 100 Mbps; 24 Workstations.

5 minutes wait after the first iteration

Test 19

Data Server; Code Server SP1; 10 Mbps; 24 Workstations.

15 minutes wait after first Iteration; Offline Folders

Test Plan

The following table summarizes the characteristics that were tested in each of the 19 tests.

Table A.2 Test Plan Summary

catap02

Hardware and Software Details for Tests

This section contains a detailed description of hardware and software used in the scalability tests.

Table A.3 Hardware

Active Directory, DNS, and DHCP Servers - Lorient, Quiberon, StMalo

Vendor: Supermicro

2 x Intel Xeon DP Processors, 2.2 GHz

4 GB RAM

100 Mbit/s Network Adapter

2x Seagate ST318406LC SCSI HD, 18GB, 10krpm

Data Server - Bourges:

Vendor: Supermicro

2 x Intel Xeon DP Processors, 2.2 GHz

4 GB RAM

1 Gbit/s Network Adapter

Adaptec 2000S SCSI RAID Controller

6x Seagate ST318406LC SCSI HD, 18GB, 10krpm (2x RAID0, 4x RAID5)

Code Server - Vichy & Strabourg

Vendor: Intel

4 x Intel Xeon MP Processors, 2.0 GHz

2 GB RAM

1 Gbit/s Network Adapter

ICP Vortex GDT8523RZ RAID Controller

5x Seagate ST318404LC SCSI HD, 18GB, 10krpm (1x single, 4x RAID5)

Workstations WS01 to WS24

Vendor: HP, Workstation x2100

Intel Pentium 4 processor, 2.6 Ghz

512 MB RAM

100 Mbit/s Network Adapter

1x Maxtor 6L020J1 IDE HD, 20GB, 7.2krpm

ATI Fire GL 8800 Graphic Adapter

Switch - Intel Express 510T

24 ports 100 BASE-T

1 port Fiber GBIC

Switch - Intel Netstructure 480T Routing Switch

12 ports 100 / 1000 BASE-T

4 ports Fiber (SX, LX, LH, GBIC)

Software

Windows 2000 Advanced Server SP3

Windows 2000 Professional Edition SP3 for Workstations

CATIA V5R9 SP1

LUM 4.6.4

Symantec Ghost 7.5

Test Results and Analysis

The following subheading discuss the test results and perform analysis on the results.

Key Statistics for Analysis

Key counters were captured during test execution. These parameters were used for analyzing the results and arriving at conclusions. Some of them and their objectives are listed below:

  • CATIA Start Time. Start time of CATIA is the most important statistic for analysis. CATIA start time is measured as the time required for starting CATIA on a workstation, either with local code or Code Serving. By comparing the CATIA start times, one can determine how scalability varies with different Code Serving scenarios and when the code is installed locally. Comparison of Code Serving with and without Offline Folders shows how this feature benefits performance. CATIA start times on 10 Mbps and 100 Mbps networks were also compared because some corporations are currently utilizing 10 Mbps network configurations. Statistics were captured for multiple iterations without a restart of the workstation to measure the effect of disk, memory, and any other caching on the start times. The minimum and maximum CATIA start times and average times with client side caching were computed to indicate the best and worst performance.

  • CATIA Total Time for Changing Workbench and Drawing. The CATIA total time value captures the total time for a user to change from one workbench to another and create a new drawing. This statistic indicates the performance of CATIA in a scenario where a user changes to a drawing workbench of CATIA and creates a drawing, resulting in an effect on both network and CPU of the local workstation.

  • Memory Utilization. The performance of CAD workstations depends heavily on the amount of memory available on the workstation and the servers. Parameters required to decide on an optimal memory configuration were captured, including Memory Used and Memory Available.

The values of the parameters mentioned above are represented graphically for all the tests in the following section.

Some of the other performance parameters measured in the tests for example, CATIA Open Time, Percentage Processor Use, and Bytes Sent and Received by Code Servers are not represented graphically in the following section, but have been used in the analysis.

Scalability Test Results

Table A.4 Effect of Local Code Vs Code Server Vs Offline Folders (for 1 Client)

Test No.

Configuration

Test 01

Local Code Local Data

Test 03

Code Server, Data Server without Offline Folder

Test 04

Code Server, Data Server with Offline Folder

Figure A: .2 Start Time

Figure A: .2 Start Time

Figure A: .3 Change Workbench Time

Figure A: .3 Change Workbench Time

Figure A: .4 Memory Usage

Figure A: .4 Memory Usage
  • Test 03 shows that with a single client, the Code Serving works well. It gives the average CATIA Start Time of below 10 seconds and Change Workbench Time of about 3 seconds, which are not too high compared with the results obtained in local code test (Test 01).

  • In Test 03, at the Code Server, the Percentage Processor Use value is always below 29%, with an average of below 2%.

  • In Test 04, the Code Serving for a single client is supplemented by the use of Offline Folders. The average CATIA Start Time is almost the same as that obtained in the previous test without Offline Folders.

  • The Change Workbench Time reduced from 3.3 sec in Test 03 to 2.9 sec in Test 04. Also, there is a small drop in CATIA Open Time from 16 sec in Test 03 to 14 sec in Test 04.

  • When we compare the performance at the Code Server in Test 04 with the same in Test 03, the Memory Used decreased by 10% and the Percentage Processor Use decreased by 2%. This shows that the performance is significantly improved by using Offline Folders with Code Serving.

  • Bytes Sent decreased by 60% from Test 03 to Test 04. Bytes received decreased by 30% from Test 03 to Test 04.

Table A.5 Effect of Code Server Vs Offline Folders (for 10 and 24 Clients)

Test 05

Code Server, Data Server without Offline Folder, 10 workstations

Test 06

Code Server, Data Server with Offline Folder, 10 workstations

Test 07

Code Server, Data Server without Offline Folder, 24 workstations

Test 08

Code Server, Data Server with Offline Folder, 24 workstations

Figure A: .5 Start Time

Figure A: .5 Start Time

Figure A: .6 Change Workbench Time

Figure A: .6 Change Workbench Time

Figure A: .7 Memory Usage

Figure A: .7 Memory Usage
  • In Test 06, Code Serving was supplemented with Offline Folders and the test was carried out for the same number of clients as in Test 05, that is, 10 clients. The average CATIA Start Time and average Change Workbench Time show slight reductions.

  • CATIA Open Time, not shown here, reduced significantly from 19.90 seconds in Test 05 to 15.75 seconds in Test 06.

  • The performance parameters (Memory Used at the Server and the Percent Usage of Processor) show substantial reduction from Test 05, in which Offline Folders were not used. Bytes Sent decreased by 60% and Bytes Received decreased approximately 30%. Thus Code Serving with Offline Folders has improved the server performance.

  • The aim of Test was to see the effect of enabling Offline Folders for Code Serving with a large number of clients (24). The average CATIA Start Time shows some reduction (approximately 10%) and average Change Workbench Time shows some increase (approximately 10%), but CATIA Open Time reduced from 20.17 seconds in Test 07 to 18.68 seconds in Test 08.

  • In Test 08, Memory Used at the Server went down by about 10%. The Percent Usage of Processor shows about a 25% reduction from Test 07, in which Offline Folders were not used. Bytes Sent decreased by more than 60% and Bytes Received decreased by approximately 23%. Thus, the Code Serving with Offline Folders has improved the server performance, even with more workstations.

Table A.6 Effect of Number of Clients on Code Server with and without Offline Folders Enabled

Test 03

Code Server, Data Server without Offline Folder, 1 workstation

Test 05

Code Server, Data Server without Offline Folder, 10 workstations

Test 07

Code Server, Data Server without Offline Folder, 24 workstations

Test 04

Code Server, Data Server with Offline Folder, 1 workstation

Test 06

Code Server, Data Server with Offline Folder, 10 workstations

Test 08

Code Server, Data Server with Offline Folder, 24 workstations

Figure A: .8 Start Time

Figure A: .8 Start Time

Figure A: .9 Change Workbench Time

Figure A: .9 Change Workbench Time

Figure A: .10 Memory Usage

Figure A: .10 Memory Usage
  • The comparative graphs show the effect of an increasing number of client workstations with and without enabling Offline Folders for Code Serving. When the results of Test 03, 05, and 07 are compared, the average CATIA Start Times show significant increases approximately 9 seconds, 11 seconds, and 20 seconds. The average Change Workbench and Drawing Times show steady increases approximately 19 seconds, 24 seconds, and 27 seconds. However, the Memory Used remains almost constant between 310 and 340 MB.

  • In Test 04, 06, and 08, the CATIA Timings do not show significant changes, but the performance parameters certainly improve due to the use of the Offline Folders. In Test 08, in which Offline Folders are enabled, the Memory Used at the Server goes down by about 10%. The Percent Usage of Processor (not illustrated in the graph) shows about a 25% reduction from Test 07, while Bytes Sent goes down by more than 60%, and Bytes Received decreases by approximately 23%. As you can see, when number of workstations increases, Code Serving with Offline Folders provides good performance.

Table A.7 Effect of Number of Code Servers with and without DFS

Test 05

Code Server, Data Server without DFS, 10 workstations

Test 09

Code Server, Data Server with DFS, 10 workstations

Test 07

Code Server, Data Server without DFS, 24 workstations

Test 10

Code Server, Data Server with DFS, 24 workstations

Figure A: .11 Start Time

Figure A: .11 Start Time

Figure A: .12 Change Workbench Time

Figure A: .12 Change Workbench Time

Figure A: .13 Memory Usage

Figure A: .13 Memory Usage
  • In Test 09, there are 2 Code Servers with DFS to serve 10 clients. The average CATIA Start Time decreased by about 15% when compared with Test 05, which used the same number of clients with a single Code Server. The average Change Workbench Time also showed a slight reduction. CATIA Open Time (not shown here) reduced from 19.90 seconds in Test 05 to 17.99 seconds in Test 09.

  • Observing the performance parameters of both the Code Servers, it is clear that both are sharing the load of Code Serving, but not always equally. The Percent Usage of Processor (not shown here), and consequently Bytes Sent and Received, are different in both the Code Servers and the processor of the Vichy server seems to be taking more load. However, Memory Used at the Server does not seem to be significantly different in both the servers.

  • Test 10 was done with two Code Servers (Vichy and Strabourgh) with DFS to see the effect of a larger number of clients (24). The average CATIA Start Time decreased by about 25% when compared with Test 07, in which the same number of clients used a single Code Server. The Change Workbench Time decreased more than 10%. Also, CATIA Open Time (not shown here) decreased marginally from 22 seconds in Test 07 to 21 seconds in Test 10.

  • The performance parameters of both the Code Servers show that both are sharing the load of Code Serving, but not equally. The Percent Usage of Processor (and, consequently, Bytes Sent and Received) differ between Code Servers and the processor of the Vichy server is again taking more load. Memory Used at the Server does not seem to be significantly different in both the servers, as earlier.

  • These tests show that Code Serving is possible with multiple Code Servers through DFS and provides satisfactory performance, even with a large number of clients.

Table A.8 Effect of Limited RAM in Client Workstations (256 MB Vs 512 MB with 1 Client)

Test 03

Code Server, Data Server with 512 MB RAM, 1 workstation

Test 15

Code Server, Data Server with 256 MB RAM, 1 workstation

Figure A: .14 Start Time

Figure A: .14 Start Time

Figure A: .15 Change Workbench Time

Figure A: .15 Change Workbench Time

Figure A: .16 Memory Usage

Figure A: .16 Memory Usage

Test 15 was done with reduced memory in the workstation (256MB), a single client, a single Code Server, and a single data server on a 100 Mbps network. CATIA Start Timings are up to approximately 30% higher with 256 MB RAM than those in a similar test configuration with 512 MB RAM (that is, Test 03). Server Utilization is similar to Test 03. This implies that for a CATIA model workstation size of approximately 18 MB, 256 MB of RAM provides a reasonable performance.

Table A.9 Effect of Number of CPUs Effect of Number of CPUs in Code Servers (1, 2 and 4 CPUs with 24 Clients)

Test 07

Code Server, Data Server, 24 workstations, 1 CPU

Test 16

Code Server, Data Server, 24 workstations, 2 CPUs

Test 17

Code Server, Data Server, 24 workstations, 4 CPUs

Figure A: .17 Start Time

Figure A: .17 Start Time

Figure A: .18 Change Workbench Time

Figure A: .18 Change Workbench Time

Figure A: .19 Memory Usage

Figure A: .19 Memory Usage
  • Test 16 used a data server, a single Code Server with 2 CPUs, 24 clients, and Offline Folders disabled. Test 17 was the same, but it had 4 CPUs in the Code Server. These tests can be compared with Test 07, in which only one CPU was installed in the Code Server.

  • In these tests, the CATIA performance parameters showed negligible differences. This implies that the number of CPUs in the Code Server does not affect the performance of CATIA, because the network is the major limiting factor. Server performance was also similar.

Table A.10 Effect of Network Speed (with 10 and 24 Clients)

Test 05

Code Server, Data Server 10 workstations, 100 Mbps

Test 11

Code Server, Data Server 10 workstations, 10 Mbps

Test 07

Code Server, Data Serve, 24 workstations, 100 Mbps

Test 13

Code Server, Data Server, 24 workstations, 10 Mbps

Figure A: .20 Start Time

Figure A: .20 Start Time

Figure A: .21 Change Workbench Time

Figure A: .21 Change Workbench Time

Figure A: .22 Memory Usage

Figure A: .22 Memory Usage
  • In Test 11, there was a data server, a single CATIA Code Server, a 10 Mbps network, 10 client machines, and Offline Folders disabled. The maximum CATIA Start Time is 43 seconds, which is close to the typical CATIA Start Time with local installation (40 seconds), but much higher than that with a 100 Mbps network (11 seconds in Test 05).

  • In Test 13, the number of clients was increased to 24. When compared with the results of Test 11, in which there were 10 client machines, CATIA performance variation is negligible.

Analysis of Network Utilization in Tests

One key observation during the tests is that the Backbone Network Bandwidth, that is, the network connectivity between the clients and the switch, saturated when more clients were accessing the Code Server. The performance data for the following tests are considered to analyze the effect of more clients on the network utilization:

  • Tests 03, 05 and 07

    Configuration: Code server, data server, 100 Mbps network: For 1, 10, and 24 clients.

    Figure A: .23 Bytes Sent per Second

    Figure A: .23 Bytes Sent per Second

    Observations. The graph clearly shows that when the number of machines is increased from 1 to 10, there is an approximately 8-fold increase in the Maximum Bytes Sent from the server. However when the number of machines is further increased from 10 to 24F, there is only a further 20% increase in the same counter. This confirms the saturation in the backbone network.

  • Tests 04, 06 and 08

    Configuration: Code server, data server, Offline Folders enabled, 100 Mbps network: For 1, 10, and 24 clients.

    Figure A: .24 Bytes Sent per Second

    Figure A: .24 Bytes Sent per Second

    Observations. Similar to Figure 3.23, this graph also indicates that increasing the number of clients from 10 to 24 is not accompanied by a corresponding increase in the Maximum Bytes Sent from the server as a result of network saturation.

Key Findings

These scalability tests simulated a variety of real life scenarios to obtain conclusive results on scalability and performance. The network bandwidth between the Code Servers and the workstations was found to be the limiting factor for Code Serving of CATIA V5 on the Windows operating environment and supporting infrastructure. Hence, binary Code Serving is an architecture capable of providing comparable performance to a local installation of CATIA, and it can also provide higher levels of availability.

Recommendations for Capacity Planning

The scalability of the CATIA Code Server and the client workstations was proved in the test environment for a 24 client configuration. All the counters which affect the performance of CATIA were captured on both server and client. Analysis of the counters on the server indicated that the peak load on the server CPU was just 20 percent. However, the network bandwidth was 100% utilized (refer to Figure 3.25). By extrapolating the optimum CPU load data, it can be concluded that the Windows server can scale up to 50 workstations in a Code Serving environment.

Figure A: .25 Bytes Sent from Code Server

Figure A: .25 Bytes Sent from Code Server

Table A.11 Server Configuration Recommendations:

Intel Xeon processors, 2.0 GHz

2 GB RAM

1 Gbit/s network adapter

ICP Vortex GDT8523RZ RAID Controller

SCSI HD, 73GB, 10krpm (1x single, 4x RAID5)

Table A.12 Workstation Configuration Recommendations:

1x Intel Pentium 4 processor, 2.6 GHz (Single Processor)

512 MB RAM

100 Mbit/s network adapter

1x Maxtor 6L020J1 IDE HD, 20GB, 7.2krpm

ATI Fire GL 8800 graphic adapter

Conclusion

The CATIA binary Code Serving solution has been performance tested and proved in the Windows architecture. This solution is recommended for CATIA deployments in large Windows technical environments. The Code Serving solution provides a better deployment paradigm resulting in reduced administration overheads in large engineering installations.

Disclaimer and Copyright Information

CATIA is a registered trademark of Dassault Systemes SA. The names of other actual companies and products mentioned herein may be the trademarks of their respective owners.

Did you find this helpful?
(1500 characters remaining)
Thank you for your feedback
Show:
© 2014 Microsoft. All rights reserved.