Test methodology and scenarios

Edit online

This section is an overview of how the WebSphere® Application Servers are configured to compare vertical JVM stacking to horizontal JVM stacking. The test scenarios are also described.

These specific attributes and parameters were used in the test methodology:
  • The WebSphere binaries are loaded from a shared DCSS mounted with the -xip option.
  • Alternatively, WebSphere binaries are loaded from a shared minidisk.
  • The physical resources in terms of processors, memory, network cards, and so on are kept constant throughout the test.
  • The number of JVMs in the LPAR are kept constant throughout the test. The number of JVMs represents the total workload that the customer needs to run.
  • The workload configuration per JVM is kept constant, which means the load level created from the workload driver is the same in all cases.
  • These parameters are varied:
    • The number of guests
    • The number of virtual CPUs (VCPUs) per guest, which varies the total number of virtual CPUs in use. The number of virtual CPUs per guest followed the rule that no guest should have more virtual CPUs than the z/VM® has physical CPUs.
    • The distribution of the JVMs among the guests.
  • The WebSphere parameters were adjusted so that no swap space is used on any guest configuration.
  • The following fields were set in all the WebSphere Application Servers:
    
WebSphere configuration:
-----------------------

   Enforce Java2 Security:      false

Servers:
   server1

 EJB/ORB ----------------------------------------
   NoLocalCopies:               true
 Web --------------------------------------------
   Min WebContainer Pool Size:  20
   Max WebContainer Pool Size:  20
 JVM --------------------------------------------
   Min JVM Heap Size:           700
   Max JVM Heap Size:           700
   Verbose GC:                  true
   Generic JVM Arguments:

 Logging ----------------------------------------
   System Log Rollover Type:    NONE
   Trace Specification:         *=all=disabled
   Rollover Size:               100
   Max Backup Files:            10
 Misc -------------------------------------------
   Enable PMI Service:          false

   Uninstall default apps:      true

Test scenarios

Six test scenarios were used, two of which were for setup parameters:
Test scenario 1: Guest scaling
Scale the number of guests: 2, 4, 10, 20, 50, 100, 200, and distribute the JVMs as described in Table 1.
Table 1. Test scenario 1 - Guest scaling: Test case configurations
Number of JVMs per guest Number of guests Number of virtual CPUs per guest Total number of virtual CPUs Ratio of virtual to real CPUs Number of JVMs per virtual CPU Guest memory size in GB Total virtual memory size in GB Comments
200 1 24 24 1.0:1 8.3 200 200  
100 2 12 24 1.0:1 8.3 100 200  
50 4 6 24 1.0:1 8.3 50 200  
20 10 3 30 1.3:1 6.7 20 200 CPU over commitment
10 20 2 40 1.7:1 5.0 10 200
4 50 1 50 2.1:1 4.0 4 200 Uniprocessor and CPU over commitment
2 100 1 100 4.2:1 2.0 2 200
1 200 1 200 8.3:1 1.0 1 200

The results of this test can be found in Test scenario 1: Guest scaling.

Test scenario 2: Varying the number of virtual CPUs for 20 guests
A CPU scaling was done for the 20 guest scenario with 10 JVM per guest.

The results of this test can be found in Test scenario 2: Varying the number of virtual CPUs for 20 guests.

Test scenario 3: Varying the number of virtual CPUs for 200 guests
A CPU scaling was also done for the 200 guest scenario with one JVM per guest.

The results of this test can be found in Test scenario 3: Varying the number of virtual CPUs for 200 guests.

Test scenario 4: Virtual CPU scaling and WebSphere threads
The number of virtual CPUs was varied to values of: 1, 2, 3, and 24.

The results of this test can be found in Test scenario 4: Virtual CPU scaling and WebSphere threads.

Setup test 1: Large guests
Define a single guest with 200 WebSphere Application Server nodes.

The results of this test can be found in Setup test 1: Large guests.

Setup test 2: Using a shared minidisk for WebSphere binaries with and without MDC
  • Replace the DCSS within the WAS installation tree by a shared read-only disk, enabled for MiniDiskCache.
  • Repeat the test without MiniDiskCache.

The results of this test can be found in Setup test 2: Using a shared minidisk for WebSphere binaries with and without MDC.