White Paper
Key Performance Metrics Exposed in EdgeSight for XenApp 5.0 and EdgeSight for Endpoints 5.0 EdgeSight Archtectural Overview EdgeSight for XenApp is implemented as an agent based solution for monitoring user experience with the XenApp environment. EdgeSight for Endpoints is also implemented as an agent based solution, monitoring the experience a user has with their Windows based desktop or laptop. In a XenApp environment, the two monitoring solutions can be combined to provide a full end to end solution for monitoring the performance of virtualized and locally installed applications. The agents collect a wide range of performance metrics; the applications users are running, the networks accessed, and the session environment those applications are contained within. That performance information is stored in a local agent database, and periodically uploaded to the associated EdgeSight server under the instruction of the EdgeSight server. The EdgeSight server provides a web based user interface from which both configuration and reporting can be performed. All data described in this paper is accessed from within the EdgeSight server console.
Critical XenApp User Performance Metrics EdgeSight for XenApp automatically collects session startup information for both the server and client portions of sessions that users establish. In addition to the startup information, client usage of the ICA channel between the client device and the XenApp server is also recorded. Both types of information are available in real-time to the session, near real-time for sessions whose information has yet to be uploaded to the EdgeSight server, and in an historical format for trending analysis. Note: EdgeSight for XenApp is available for Presentation Server versions 4.0 and higher. Please reference the EdgeSight for XenApp documentation for information on what data is available from each version of Citrix Presentation Server. This document assumes that the EdgeSight for XenApp agent version is 5.0.
2
Session Startup Information Note: Client side session creation information is only available when the client connection is established. If a client connection is already established, ICA will use that connection for the new server session. Note: Server side session creation information will not appear if session sharing is enabled. The following items are core in understanding session creation in a XenApp environment. •
Server Startup Credentials Obtention is the length of time required for the XenApp server to query the user for their credentials. In general, this value is significant for published desktops where the user is presented a secondary login or conditions of usage and corresponds to the length of time the user spent entering this information.
•
Device Mapping is the length of time spent on the XenApp server mapping the users’ devices to the XenApp environment. Users with a large number of local devices will see a larger value for this item, and user experience optimizations can be made by disabling device mappings on the XenApp farm which are not appropriate to the published application.
•
Printer Creation is the length of time spent mapping the users’ local printers to the XenApp session. If printer mapping is configured to be asynchronous, this value will be zero as asynchronous printer mapping occurs after the session has been created.
•
Profile Load is the length of time spent loading the users’ server profile. Larger profiles will take longer to load, and the overall XenApp server resource utilization will impact profile load.
•
Login Script Execution is the length of time spent running the login script associated with the user for the given application. Performance improvements can be made by customizing the login script to both the user and the application.
•
Launch Page Web Server Duration is the length of time associated with the launching of the web page requesting the ICA file from the XenApp server. A large value indicates the Web Interface server is experiencing performance issues.
•
Client Startup Session Creation is a measure of how quickly the ICA client launches on the client system. A large value indicates that the client machine is experiencing performance problems launching the ICA client.
•
ICA File Download is a measure of how long it takes for the ICA file describing the published application or desktop to be downloaded to the client. A large value indicates network performance issues may exist between the client and the XenApp farm.
ICA Performance Within the EdgeSight environment, ICA performance is recorded both as a generic network statistic, and at the virtual channel level. Within the virtual channel data, it is important to reference the overall ICA volume, and the available ICA bandwidth.
3
Note: EdgeSight for XenApp collects information from all the standard Citrix ICA channels, and presents the critical channel information both in real-time and historical reports. Custom channel traffic and non-ICA virtual channel information is not collected. The following items are core in understanding user experience with a session. •
ICA network latency is the network latency between the client and the server. A high value will be perceived by users as a lack of responsiveness with their application.
•
ICA Session Compression is an indication of the efficiency of the ICA compression on the specific data being transported.
•
ICA data volume is the volume of data being transported over all ICA virtual channels, including any custom channels. If the data volume is large, and users are experiencing performance problems, it is valuable to investigate which channel is dominating the ICA connection, and if the data volume in the specific channel is appropriate both to the application and the users’ connection.
•
ICA printer data is the volume of data being transported over the ICA channel for printer operations. A non-zero value indicates that a print operation is occurring.
•
ICA drive data is the volume of data being transported over the ICA channel for file copy operations. A non-zero value indicates that file copies are occurring. It is important to note that depending on how drive mapping is performed, and application usage, some applications may access the users’ local drive as a temporary file location. Such behavior can increase the data volume on the drive data virtual channel, and could create a scenario where user experience is impacted.
•
ICA SpeedScreen data is the volume of video data being transported over the ICA channel dedicated to SpeedScreen operations. If SpeedScreen is disabled for a given application, this value will be zero. In such scenarios, video data will be transmitted over the ThinWire channel.
•
ICA ThinWire data is the volume of video data being transported over the ICA channel dedicated to ThinWire operations. If SpeedScreen is enabled for a given application, this value will be zero and the video data will be transmitted over the SpeedScreen channel.
•
ICA clipboard data is the volume of data being sent to or received from the published application. A high value indicates that heavy clipboard use is being performed, and it’s important to be aware that if binary or formatted data is being transferred via the clipboard that clipboard data is correspondingly increased.
Critical System Performance Metrics EdgeSight natively monitors the performance of the computer on which it is running. The system performance metrics collected are identical for both the XenApp and Endpoint environments, though the meaning of a given value may vary whether EdgeSight is installed on XenApp or Endpoint. This information is available via the various real-time charts, and once uploaded to the EdgeSight server is also available for historical trending and analysis.
4
•
Total Processor Time is typically referred to as CPU percent, and is normalized to the number of processor cores as seen by Windows. Due to this normalization, interpretation of the data does not require knowledge of the physical characteristics of the computer.
•
Privileged Time measures the amount of CPU time spent within core operating system components and drivers and is valuable in understanding overall CPU workloads. As a general rule, if privileged CPU resources are a significant portion of the overall CPU usage, then application performance can be impacted, particularly in XenApp environments where a large number of users are on the XenApp server.
•
Processor Queue Length is the number of work items scheduled for CPU execution. If the number is non-zero, then this means that some work items are not being executed as they become available, and application responsiveness will be impacted. As a general rule, work items become queued if either the CPU is unavailable due to excessive CPU utilization, or if the CPU is unavailable due to resource contention.
•
Current Disk Queue Length is the number of disk accesses which are waiting to be executed. If this value is non-zero, then the disk is becoming a bottleneck to overall system performance.
•
Available Kbytes is the available system memory expressed in kilobytes.
•
Active and Total Sessions are valuable in a XenApp environment to gauge both the number of users on a XenApp server, and the number of users who are experiencing any performance degradation.
•
Failed and Reset Connections indicate the overall stability of the network environment between the computer and the remote devices. Any non-zero value warrants investigation.
•
Average Network Errors indicates the overall network experience with the network services accessed by applications running on the computer. Network errors can come from both network related issues, such as a timeout, and from user input or configuration errors, such as an invalid hostname.
•
Average Client, Network and Server Delay indicate the magnitude of delay associated with network connections from the computer to the network resource, and whether the delay was dominated by the client, network or server portion of the transaction.
Critical Application Performance Metrics EdgeSight natively monitors the application performance and resource utilization for all applications running on a computer system, regardless of the type of application. No modification of an application is required for this monitoring to occur. The application performance metrics collected are identical whether the application is running on an Endpoint, or within a XenApp session. Note: Real-time charts will only show a selected sub-set of the available application and network performance data. Detailed analysis can be performed using either the Excel based real-time reports, or the historical reports.
5
•
Total Processor Time is typically referred to as CPU percent, and is normalized to the number of processor cores as seen by Windows. Due to this normalization, interpretation of the data does not require knowledge of the physical characteristics of the computer.
•
Privileged Time measures the amount of CPU time spent within core operating system components and drivers and is valuable in understanding overall CPU workloads. As a general rule, if privileged CPU resources are a significant portion of the overall CPU usage for a process, that process may appear unresponsive or sluggish to the user.
•
Thread Queue Length measures the number of work items waiting for threads to execute them. A high thread queue length generally will correspond to a period of time when a user perceives the application as being unresponsive or sluggish. It’s important to note that a high thread queue length for an application may mean that the application is waiting on system resources, but can also mean that a third party application is causing the performance slowdown. A good example of the latter scenario is an anti-malware solution which operates by either causing one of their DLLs to load in the application, or by modifying disk access routines, to scan files prior to their being opened. During that scan, the application will be blocked waiting on the outcome of the scan and depending upon application architecture the thread queue length may grow and the application may be deemed unresponsive by users.
•
Working Set is the amount of memory currently allocated to the process.
•
Network Delay on a per host, protocol or port basis is only available if the application is network enabled, but if it is the data provides a breakdown of the TCP performance experienced by the application.
Conclusion EdgeSight for XenApp 5.0 and EdgeSight for Endpoints 5.0 natively collect a variety of system, application, network and XenApp performance metrics. This performance information is invaluable in understanding and diagnosing performance problems reported by end users. While EdgeSight collects much more information than was presented in this paper, the key metrics described form the basis for many analysis scenarios.
