We're running three Windows Server 2025 RDS session hosts with FSLogix profiles stored on an SMB share. Each host has 20 vCPUs, 96 GB of RAM, and around 12–15 concurrent users. Outlook is running in Cached Exchange Mode, and one host also has an NVIDIA Tesla T4 passed through for GPU acceleration.
The monitoring data looks normal: CPU and memory usage are low, GPU utilization is minimal, storage I/O and IOPS look healthy, and SMB/profile-share latency is generally below 1 ms. The environment uses 10 Gbit networking, and the storage sometimes reports latency in the microsecond range.
Despite that, users report that applications feel sluggish. Explorer and Outlook can become sticky, and the entire RDS session sometimes feels choppy even though there is no obvious CPU, memory, disk, network, or GPU bottleneck. Our older Server 2022 hosts actually feel more responsive, even when handling more users.
The virtual machines run on Proxmox 9. The FSLogix profile storage is on an additional SSD RAID 5 volume. We have excluded the SMB profile share and the recommended file types and paths from our antivirus configuration.
Has anyone run into similar behavior with FSLogix on Server 2025? I'm particularly interested in cases where standard performance counters looked fine but users still experienced intermittent delays. What ultimately turned out to be the cause?
4 Answers
As a baseline, try a smaller host configuration with fewer vCPUs and compare the experience. Some workloads can feel worse with excessive vCPU allocation if the hypervisor has to schedule larger virtual machines. Also test without GPU redirection and with a plain session to separate an RDS graphics issue from an FSLogix or storage issue.
Antivirus or EDR is still a strong suspect, even when the share and disks look idle. Make sure the exclusions cover the FSLogix executables, profile-container locations, mounted VHD or VHDX files, and the relevant temporary and user-profile paths. A security scan can introduce small delays that disappear in average latency graphs but are very noticeable inside an interactive session.
The hosts may be overprovisioned for this user count, so the headline CPU and RAM numbers don’t rule out virtualization-related latency. Check the Proxmox host for CPU steal or ready time, host contention, NUMA placement, storage wait, and scheduling delays. Also review the vCPU-to-physical-core ratio and page-file activity rather than only looking at guest utilization.
First, test a few users with FSLogix disabled. That quickly tells you whether the delay is coming from profile-container operations or from the RDS hosts themselves. Compare the results with the older Server 2022 setup, including the exact FSLogix, Windows, and security-product configuration—not just the storage metrics.
That’s a good isolation test. I’ll compare users with and without FSLogix and check the configuration differences between the 2022 and 2025 hosts.
The SMB share and common file types are already excluded, but I’d verify the exclusions against the complete vendor guidance and check whether the security agent is logging scans or blocked operations on the affected hosts.