About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
Measure before you choose. From each office, during business hours, record the round-trip time to each candidate hosting location, then compare it with the way your staff work. For staff using remote desktops for everyday office work, Microsoft's figures give a useful yardstick: under 150 milliseconds is the target, and above 200 milliseconds users may notice. Work involving rendering or video, such as design, imaging or CAD, needs better than that. For desktop applications that talk directly to a hosted database, our view is that the acceptable delay is much lower again, because those applications can make many round trips for one action. Choose the location that serves the offices where most work happens, and put chatty applications next to their database.
Why it matters
Distance adds delay that no amount of bandwidth removes. A business with staff in two cities can pick a location that suits one office and frustrates the other, or run an accounting package that feels instant in head office and sluggish everywhere else. Measuring first costs an afternoon; moving servers later costs a project.
How it works
Round-trip time is the figure to measure
Round-trip time (RTT) is how long a small message takes to reach the server and come back. Windows can measure it: PowerShell's Test-NetConnection "Displays diagnostic information for a connection", and its output includes a "PingReplyDetails (RTT)" line. Ask each candidate provider for a test address in each location.
Reference points for remote desktops
Microsoft's guidance for its own remote desktop service gives the clearest public thresholds. Its prerequisites say: "Round-trip time (RTT) latency from the client's network to the Azure region that contains the host pools should be less than 150 ms." Its troubleshooting guide is more specific about the kind of work: "latency up to 150 ms shouldn't affect user experience that doesn't involve rendering or video", and "Latencies between 150 ms and 200 ms should be fine for text processing. Latency above 200 ms might affect user experience."
Some users need much better. Microsoft's Remote Desktop network guidelines note that assistive technology such as Narrator needs "a connection round trip time (RTT) of 20 milliseconds (ms) or better for the best user experience". If any staff rely on a screen reader, measure for them specifically. Staff who do design, imaging, CAD or video work through a remote desktop also need a shorter delay than the 150 ms figure; measure and test with their real work.
Why desktop applications are different
With a remote desktop, the screen travels over the distance and the application talks to the database inside the hosting location. With a desktop application installed on each PC and its database hosted elsewhere, every query crosses the distance, and when one action makes a series of queries one after another, the delays add up. Our companion article on splitting an application between office and cloud works through the arithmetic.
What to choose and why
A measurement plan
- List the offices and the number of staff at each.
- Get test addresses for each candidate location from each provider.
- Measure at business hours on several days, from a wired PC at each office, taking many readings rather than one. Record the middle and the worst figure. Many firewalls and servers drop ping. Test-NetConnection's "-Port" option then confirms whether a TCP connection to a port the provider names succeeds, but that is a connectivity check, not a timing: it does not time the TCP connection. For timing where ping is blocked, use the application's or remote desktop client's own connection statistics, or a TCP timing tool your provider supports. Our published article Proving packet loss and latency to a provider: reading mtr and pathping correctly covers longer, repeatable measurements with MTR and PathPing.
- Note how each group works: remote desktop, web application, or desktop application with a hosted database.
- Compare with the reference points above.
Choosing
| Situation | Better choice |
|---|---|
| One main office, small branches | Host nearest the main office if branches still meet the remote desktop target |
| Two large offices, similar size | Host where both meet the remote desktop target; move desktop applications onto remote desktops |
| Chatty desktop application used from several offices | Run it on a remote desktop beside its database, not on each PC |
| Staff using assistive technology | Measure their connection; choose for them first |
The table is editorial guidance from the reference points; your measurements decide.
Illustrative example
A firm has 30 staff in one capital city and 12 in another. Both offices measure under 40 milliseconds to either candidate location, well inside the remote desktop target. But the firm's practice-management application is installed on each PC and queries its database directly. The firm hosts in the larger office's city and moves the application onto a hosted remote desktop, so the second office no longer feels the distance on every query. The example is illustrative.
Read next
Related articles you may want next.
- HostingSplitting an application between office and cloud: why chatty database traffic breaks itWhy a desktop business application can slow badly when only its database moves to a hosted server while the program stays on office PCs, how to count the round trips before deciding, and the placement choices that avoid the problem.Read article
- PerformanceProving Packet Loss and Latency to a Provider: Reading mtr and pathping CorrectlySlow or lossy link and the provider wants evidence? Read mtr and pathping correctly, ignore ICMP rate-limit artefacts, and build a time-stamped evidence pack.Read article
- HostingSizing an RDS farm for mixed workloads and one host out of serviceA worked, explicitly hypothetical capacity worksheet for mixed RDS users, private and shared memory, morning sign-ins and the loss of one session host for maintenance.Read article
Risks and limits
- Measurements change. Office internet links, provider routes and congestion vary; re-measure when an office changes provider.
- Thresholds are Microsoft's for its own service. They are useful yardsticks for remote desktops generally, not guarantees for other products.
- Packet loss matters too. Microsoft's network guidelines say its recommendations "apply to networks with less than 0.1% loss". A short RTT on a lossy link can still feel poor.
- Location also has legal and contract aspects. Where data must stay, and which provider locations are offered, are separate questions from latency.
- Home workers count. Their home internet adds its own delay; include a sample in the measurements.
Sources
Opened 15 November 2026; the connection-quality and Test-NetConnection pages and the published articles reopened 16 November 2026.
- Microsoft Learn: Prerequisites for Azure Virtual Desktop, updated 4 November 2026.
- Microsoft Learn: Troubleshoot Azure Virtual Desktop connection quality, page shows last updated 22 January 2025 (page metadata: modified 2 October 2025).
- Microsoft Learn: Network guidelines, Remote Desktop Services, updated 15 September 2025.
- Microsoft Learn: Test-NetConnection, updated 14 May 2025.
Related articles: Proving packet loss and latency to a provider: reading mtr and pathping correctly; Sizing an RDS farm for mixed workloads and one host out of service.
Related service: Cloud hosting.
Learning notes
- Measure round-trip time from every office during business hours before choosing a location.
- For remote desktops without rendering or video work, Microsoft targets under 150 ms and says above 200 ms might affect users; design, imaging and video need better.
- Assistive technology needs 20 ms or better for the best experience.
- Desktop applications querying a distant database multiply the delay by every sequential query.
- Take many readings over several days; if ping is blocked, a TCP port test confirms reachability only, so take timings from the application's or remote desktop client's own statistics.
- Put chatty applications next to their database, for example on a hosted remote desktop.
This article was originally published on .