I am posting the Q&A for the TechTalk we had on XenApp Power and Capacity Management. It was well attended (more than 50 customers and partners) with lots of time for Q&A. Check out the webinar recording if you were not able to attend.
(You can follow me on twitter)
Q) How does a combination of XenServer with virtual XenApp work? So, is it possible to first shut down the unnecessary XenApp servers and then the unnecessary physical servers running XenServer?
A: Power and Capacity Management does not manage the XenServers themselves. It only manages XenApp servers. So, if you have no virtual XenApp servers running on a XenServer it will still be powered on. Why? You may still have non-XenApp workloads running there like web servers or e-mail servers, etc. But you can use other management tools to power manage the hypervisor.
Q) Will it be possible to install the PCM agent in a “standard” provisioning server image?
A: Yes. Power and Capacity management works with Wake on LAN which is how provisioning server works to deliver images to bare metal. Wake on LAN is supported in addition to virtual machines for Power and Capacity Management.
Q) So, Can I provision my XenApp servers with provisioning server and the collector gets the different names of the server or is a special configuration needed?
A: Yes. When you install the agent, there is a configuration for Wake on LAN or Virtual machines. You select Wake on LAN and configure accordingly for the image you are creating.
Q) Must I also choose “Wake on LAN” if the provisioned XenApp server is a virtual server on XenServer?
A: You can have a single PVS image to work across physical or virtual deployments. Even when using PVS, if you are booting servers via XenServer, you would choose XenServer. If you are booting via Wake on LAN on the physical hardware then you would choose Wake on LAN
Q) Can PCM and SCOM work together?
A: Out of the box, we are not planning any integration with operations manager. However, everything being done through the GUI can be done through WMI, so this can be customized.
Q) Should the concentrator be installed on a XenApp server?
A: No. You do not want to install this on a XenApp server that hosts users sessions. You won’t be able to power manage it if you do. Also, concentrators can manage servers across farms so you want a farm-independent server if you can do so.
Q) Instead of number of sessions, is there a way to set the Capacity by percentage of Server resource utilization e.g. CPU, RAM, Page File Usage?
A: Yes. We are enabling you to use nominal capacity (sessions) or load evaluator and have this all calculated as a percentage of load evaluator. This will be enabled in the final release but is not available in the technology preview.
Q) In terms of new session distribution, what takes precedence, the XenApp load evaluator or the Power and Capacity Management evaluator?
A: Power and Capacity Management takes precedence for server power on/off and consolidation only. For load distribution, the XenApp load evaluators are used.
Note that PCM adjusts the load evaluation for servers that are not selected to receive sessions. PCM will keep up to “Minimum Available Servers” available for logon. Other servers’ load evaluators will be set to 20,000 (you will see that if you run QFARM /load).
Q) I am wondering how server monitoring will have to be modified to allow for servers “shutting down” and suppressing alerts. Do you have any examples of solutions?
A: EdgeSight already differentiates between planned and unplanned reboots. Since reboots initiated by PCM will be planned reboots, you should be able to create alerts only for unplanned reboots
Q) Can you please explain the process of draining a server. We publish the desktop so we have some users logged on for 8+ hours. Also we have shift workers.
A: PCM selects up to “Minimum Available Servers” to accept logons. All other servers have their LE values set to 20,000 and therefore won’t be selected to host apps/desktops. When an available server reaches the optimal load, the server with highest load, but under optimal load, is chosen to receive logons.
Draining is a side-effect of the model above. As shift workers start to login, they will do so on the highest loaded servers under optimal load policy. As the previous shift logoff, their servers will automatically drain until empty. The new shift will only consume enough servers to meet its demand.
Q) If you provision server through PVS with the agent already installed and configured for a workload called “office 2003”, if you want to change the workload to “office 2007” would you need to change or create a new PVS image to reflect this change?
A: Workloads can be assigned via GPO, so it’s possible to use the same PVS image for more than one PCM workload. In the example above, if Office was virtualized via App-V or Streaming, then the same image could be used.