The impact of Mirage activities on the end-user experience depends on several factors.
The following factors can impact the user experience on Mirage activities:
Initial load and utilization of the system
Resource boundaries for dedicated resources such as ESX host CPU and memory
Resource boundaries for shared resources such as storage IOPS and network bandwidth
The optimal number of concurrent layer updates should take into consideration all of the above factors. As a result, the absolute value can vary between different environments.
The baseline load and utilization of the system provides an initial range or additional load that can be applied on certain resources to the pool. Typically, VDI systems are designed to be under 85% of CPU load allowing some room for peaks. Depending on the decision by the Mirage administrator, part of this buffer can be used to address the calculations related to layer updates.
Dedicated resources, typically CPU and memory of the ESX host, are linearly scaled. When a single ESX host can handle 20% of the updating endpoints, the whole pool can handle a similar amount of updating endpoints assuming random distribution.
When the system is finally bounded by shared resources, typically back-end storage arrays, the amount of concurrent updates reflect the overall load that is done on the storage array by all users, even when a single ESX host can handle the load.
For additional sizing considerations in View environments, see the Architecture Design Elements and Planning Guidelines in the View Architecture Planning documentation.
The following reference tests describe the potential end-user impact on a single ESX host performance using different values for concurrent layer updates. All the tests were performed using a VDI workload generator that automates and measures a typical end-user activity. The automated applications used for the tests were Microsoft Office, Adobe Reader, watching a video, and so forth. Operations used on these applications were opening a file, browsing the Web, modifying files, and saving and closing files. In addition to these activities, a defined number of endpoints performed Mirage layer updates.
Test Setup
The reference tests used the following configurations:
Single ESX host (16 cores, 196GB RAM)
SAS Storage
100 VMs
20 and 50 concurrent layer updates
Two types of loads:
Group A - CPU intensive
Group B - IO intensive
Test Results
The following table displays the reference test results for the number of endpoints concurrently performing Mirage layer updates.
Baseline |
20 Endpoints |
50 Endpoints |
|
|---|---|---|---|
Response time increase: Group A |
- |
21% |
74% |
Response time increase: Group B |
- |
17% |
83% |
