Exercise 4 Server Software Performance
Proper web service instance settings are important for leveraging available platform computing resources. This exercise will use the CPT Design tab to demonstrate the importance of proper web service configurations.
- Web services should be configured based on expected use patterns.
- Batch and geoprocessing tasks consume a core processor during execution, while lighter web services can require three to five service instances per core processor to reach peak platform throughput levels.
- The CPT Design tab can be configured to demonstrate the processing loads for each of these service configurations.
Use the CPT to evaluate server software performance
In this exercise, you will perform the following tasks:
- Configure the CPT Calculator tab for demonstrating batch service loads.
- Configure the CPT Design tab for demonstrating batch service loads.
- Configure the CPT Design tab for demonstrating map service loads.
Step 1: Demonstrate batch process service instance loads on the CPT Calculator tab
- Select the CPT Calculator tab.
- Use the Copy Sheet command to create a new Design tab copy, and label it 4.1.
Best practice: It is good practice to move each new design to a new tab and label newly created CPT tabs for future reference.
You will use the CPT Calculator to configure single-platform configuration.
- Select Software Technology Performance Factors for the following workflow: AGS REST 2D V Med 100%Dyn 13x7 PNG24
- Select DB_DBMS data source in cell J6.
- Configure a single tier platform architecture by selecting Single (cell A9) and Minimum (cell A10) in the Platform Architecture section.
- Select Xeon Gold 5122 4 core (1 chip) 3600 MHz platform in cell D10.
- Select Batch (cell J15) to move client load to GIS Server machine.
- Type 1 (cell D25) to fix ArcGIS Server nodes to one platform.
You have just installed all workflow and database software on the GIS platform server.
A batch process is created by selecting zero minimum think time (cell D6).
1. Use the CPT Calculator model to complete the following table.
- In cell A6, type the number of service instances.
Excel will complete the analysis.
The cell D6 background turns green when workflow productivity (E6) is valid.
- • Workflow throughput (TPM) is reported in cell C6 (multiply by 60 min/hr to convert to transactions per hour [TPH]).
- • Workflow response time (sec) is reported in cell K12.
- • Server utilization is reported in cell H10.
Configuration represents maximum batch process throughput with a selected number of service instances. A single batch process consumes one processor core.
Results of the test runs are summarized in the following charts.
Best practice: There is very little gain in throughput beyond one service instance per core (four total instances). The N+1 configuration (five instances) ensures maximum server throughput load. Peak load response times increase with an increasing number of service instances.
Step 2: Demonstrate batch process service instance loads on the CPT Design tab
- Select the CPT Design tab.
- Use the Copy Sheet command to create a new Design tab copy, and label it 4.2.
Best practice: It is good practice to move each new design to a new tab and label newly created CPT tabs for future reference.
You will use the Software Configuration module to configure single-platform configuration.
- Open the Software Configuration module in the CPT Design tab, columns J through R
Tip: Expand excel columns by selecting the + located above column S.
- Configure the default install so that Client, Web, SOC, and DBMS software are all executed on the GIS server tier.
- • In cells J5, L5, N5, and Q5, select GIS as the default platform tier.
You have just installed all workflow software on the GIS platform tier. All colored cells in workflow rows should be default. The data source for all workflows is SDE_DBMS (column R).
The WAN and Internet rows can be hidden for this exercise to create a simpler display.
WARNING: WHEN HIDING WORKFLOW ROWS, ENSURE THAT YOU REMOVE ANY PEAK CONCURRENT LOADS IN COLUMNS C AND D. WORKFLOW LOADS WILL BE APPLIED TO THE SERVER PLATFORMS EVEN WHEN THEY ARE HIDDEN.
- Close the Software Configuration module.
You will now use the CPT Design Platform Selection module to select platform configuration.
- In column B, select the following platform configuration for GIS: Platform Tier 09.
- • Xeon Gold 5122 4 core (1 chip) 3600 MHz
- • In column H, set GIS tier Fix Nodes to 1.
You have just forced the number of available servers equal to 1 for the GIS tier.
You will use the Requirements Analysis module setup for batch workflow loads.
- In columns C and D, select and delete any existing users or TPH throughputs (Requirements Analysis white cells).
- In cell B6, select the following workflow:
- • AGS REST 2D V Med 100%Dyn 13x7 PNG24
- Select Client (cell D4) to identify client batch processes.
- In cell D6, set peak concurrent client to 1 (single batch process).
- In cell H5, set LAN bandwidth to 10000 Mbps.
- In cell T6, select a minimum think time of zero (0).
A batch process is created by selecting zero minimum think time.
WARNING: AN ENTRY IN COLUMN T WILL OVERRIDE THE FORMULA AND TURNS THE CELL WHITE. REPLACE THE FORMULA FROM ANOTHER ROW IF YOU NEED TO RECONFIGURE THE WORKFLOW FOR A DIFFERENT ANALYSIS.
- Set the RESET function to ADJUST in cell T2, and Blink to 10 in cell U2.
2. Use the CPT Design model to complete the following table.
Set the RESET function to ADJUST in cell T2, and Blink to 10 in cell U2.
The Blink setting must be reduced to a smaller value to compute productivity during high utilization loads.
Best practice: The Blink setting should start at 10 and reduce by a factor of 10 until reaching a valid productivity.
- In cell C6, provide the number of service instances.
Excel will complete the analysis.
The cell C6 background turns green when workflow productivity (E6) is valid (make sure that platform utilization is less than 100 percent).
- • Workflow throughput (TPH) is reported in cell H6.
- • Workflow response time (sec) is reported in cell V6.
- • Server utilization is reported in column V of the platform tier.
Configuration represents maximum batch process throughput with a selected number of service instances. A single batch process consumes one processor core.
WARNING: DO NOT LEAVE THE RESET FUNCTION (CELL T2) IN THE ADJUST SETTING (ADJUST FUNCTION LEAVES EXCEL IN A CIRCULAR CALCULATION MODE). MOVE TO THE SAVE SETTING TO SAVE THE SOLUTION BEFORE LEAVING THE TAB.
Results of the test runs are summarized in the following charts.
Best practice: There is very little gain in throughput beyond one service instance per core (four total instances). The N+1 configuration (five instances) ensures maximum server throughput load. Peak load response times increase with an increasing number of service instances.
Step 3: Demonstrate web mapping service instance loads on the CPT Design tab
- Use the Copy Sheet command to create a copy of 4.2 final design, move it to the end, and label the new tab 4.3.
You will use the Requirements Analysis module setup for web service workflow loads.
- In cell C6, set the peak concurrent users to 1.
- In cell T6, enter a minimum think time of 0.01.
A minimum think time greater than zero will change the CPT Workflow model to represent web service random arrival times. A small think time (0.01 sec) will simulate pooled service instance assignment supporting random arrival of user map requests.
- In cell D1, select Test mode.
Note: Test mode forces the CPT model to calculate workflow productivity based on selected minimum think time. RESET ADJUST function (T2:U2) must be used to calculate a valid productivity. Throughput cells in column C and D will turn green when calculating a valid workflow.
3. Use the CPT Design model to complete the following table.
- Set the RESET function to ADJUST in cell T2, and Blink to 10 in cell U2.
The Blink setting must be reduced to a smaller value to resolve productivity during high-utilization loads.
- In cell D6, type the number of service instances.
Excel will complete the analysis.
The cell D6 background turns green when the workflow productivity (E6) is valid.
- • Workflow throughput (TPH) is reported in cell H6.
- • Workflow response time (sec) is reported in cell V6.
- • Server utilization is reported in column V of the platform tier.
Configuration represents the maximum web service throughput with the selected number of service instances. Optimum throughput is reached at 3–5 service instances per core (12–20 instances for a 4-core server).
There is very little gain in throughput beyond four service instances per core (16 total instances). Web service configuration with 3–5 instances per core assures maximum server throughput load. Peak load response times will increase with the increasing number of service instances.
Answers to Exercise 4 questions
Exercise 4: Use the CPT to evaluate server software performance
1. Use the CPT Calculator model to complete the following table.
2. Use the CPT Design model to complete the following table.
3. Use the CPT Design model to complete the following table.
