Implement monitoring, scaling, and optimization
Operational focus
This domain tests whether you can observe Citrix and Azure layers together, scale capacity economically, and optimize image and host performance. Citrix Monitor shows sessions, logon duration, failures, and machine registration. Azure Monitor, Log Analytics, VM insights, and diagnostic settings show platform metrics, guest performance, and security events. Autoscale and power management control cost.
Core rules and concepts
- Collect Azure platform metrics and activity logs into Log Analytics. Enable boot diagnostics and guest diagnostics where needed.
- Use Citrix Monitor for session and machine health; use Azure Monitor for CPU, memory, disk latency, network, and availability.
- Autoscale needs a schedule, minimum and maximum machines, a buffer, and a scale-out/scale-in cooldown. Keep enough spare capacity to absorb boot time.
- Power management should deallocate idle machines rather than merely stop them inside the OS. Deallocation stops compute billing.
- MCS image updates should be tested on a small catalog first. Use separate resource groups and image versions to roll back.
- Optimize by right-sizing VM SKUs, using premium storage only where needed, reducing profile bloat, and controlling egress.
Worked example
A catalog serves 100 users from 08:00 to 18:00, with 20 percent concurrency at peak. Peak demand is 20 sessions. At five sessions per host, four hosts are needed. Autoscale is set with minimum four, maximum eight, and a 20 percent buffer, so peak target is five hosts. Scale-out triggers when average sessions per host exceed four for 10 minutes; scale-in waits 30 minutes. Machines are deallocated outside business hours. The reasoning is that the buffer covers boot time and the cooldown prevents flapping.
Traps that cost marks
Setting minimum capacity too low causes boot storms and failed logons. Stopping a VM inside Windows does not stop Azure compute billing; deallocation does. Scaling only on CPU misses memory-bound session hosts. Updating the production image without a pilot catalog risks wide outages. Ignoring disk latency and profile IOPS leads to poor logon performance despite low CPU.
Key points
- Use Citrix Monitor for session health and Azure Monitor for platform metrics.
- Autoscale needs schedule, buffer, minimum, maximum, and cooldown settings.
- Deallocate idle VMs to stop compute billing.
- Pilot image updates before broad catalog rollout.
- Right-size SKUs and storage; watch disk latency, not just CPU.