Work tasks and route metrics
When choosing a VPN for remote work, start with the task that can least afford interruptions. For frequent calls, focus on packet loss and jitter during real meetings. If you mainly work in documents, check whether edits sync promptly and sign-ins stay active. For frequent code pulls, test sustained transfers too. The node latency shown in a client only reflects its test destination; it doesn’t tell you how the route performs to a meeting server or code repository.
Route selection: First confirm which connection methods your company permits, then choose a primary route based on the task that can least afford interruptions. If your company allows a direct connection and the task is more reliable that way, there’s no need to route it elsewhere.
| Work Task | What to Check First | How to Compare |
|---|---|---|
| Video calls | Packet loss in both directions, jitter, and upload performance during screen sharing | Use the same video conferencing app for a call and screen share. Compare its call statistics with what the other person hears. |
| Cloud documents | Whether saving and syncing repeatedly stall, and whether sign-in and attachments work | Edit the same approved test document from both ends. Confirm the changes appear on the other device, then check the document again after reopening it. |
| Code repositories | Whether authentication succeeds, transfers stay connected, and large files finish downloading | Use your usual HTTPS or SSH method to access an approved test repository. Record authentication and pull results separately. |
Video calls: Tell packet loss, jitter and upload congestion apart
Latency is the time data takes to travel; jitter is variation in that time. Even when average latency looks low, inconsistent delivery can make speech choppy. Packet loss means some data never arrives. Start with the call statistics in your conferencing app, since a browser speed test usually doesn’t use the same media server. If the stats separate sent and received data, match the affected direction to the symptoms: if colleagues can’t hear you clearly, check the outgoing direction first; if you can’t hear them, check the incoming direction.
If problems occur only while screen sharing or syncing files, pause background uploads and compare performance before and after. If possible, test over a wired connection to rule out wireless interference. Switch routes only if the same issue repeatedly occurs on one candidate route with the same network and meeting settings. Some conferencing apps prefer UDP for media and may use another transport when the network is restricted. Joining a meeting doesn’t prove that audio and screen sharing are using the expected route.
Shared documents and code repositories: Test the full workflow
Being able to open a document’s home page doesn’t mean real-time collaboration is working. Sign-in, edit syncing, and attachments may rely on different domains or persistent connections. For a fair comparison, use the same account and permissions, and go through the full sequence: open, edit, wait for sync, then reopen. If the other person never sees your changes, first check whether the page indicates it’s offline. Then review the client connection logs to see which rules matched sign-in and sync requests. If only attachments fail, check the file service and permissions separately rather than assuming the entire route is too slow.
For code repositories, test authentication and data transfer separately. If an error occurs before authentication completes, check your credentials, your organization’s single sign-on requirements, and your company’s access status. If the connection hangs or drops only after authentication succeeds, compare sustained transfers across routes. A small incremental pull working well doesn’t guarantee that an initial clone or large download will finish. If you use Git LFS, include a large-file download in your test, since its service address may differ from the repository’s. Use the same test file for each comparison, and account for caching, which can make a later download seem faster.
Comparing direct connections, relay routes and dedicated lines
In split-tunneling rules, “direct” usually means app traffic doesn’t go through a proxy. A product’s “direct node” usually means the client connects straight to that node; these are not the same thing. A relayed route passes through an entry point before forwarding traffic to an exit. This may improve a poorly covered segment, but congestion at the entry point can also add latency. IEPL means International Ethernet Private Line. An IEPL label doesn’t mean every segment from your home device to a work app uses that line; local access and routing from the exit to the app still affect performance.
Use region labels only to narrow down options. An app may receive requests through different entry points, and its meeting media server may be in a different region from its sign-in server. Keep the device, network, app, and task the same, and test each route during your usual work hours. Use the company-approved direct connection as a baseline, and note which step stalls each time. If two tasks favor different routes, prioritize the real-time call, which is harder to recover, and schedule retryable large-file transfers after the meeting.
Split-tunneling rules: Setup and verification
- Test the baseline first. With your organization’s approved default settings, test a call, document syncing, and repository access. Record which step fails, not just whether a page opens.
- Then narrow the rules. Change routes only for approved external apps where testing confirms an improvement. Depending on your client, rules may match apps or domains. Check rule order so broad matches don’t override more specific entries.
- Verify the actual route. Use the client’s connection logs to check which rules matched and which exit was used, then test the app itself. Connection rules and DNS resolution paths may differ. If queries for external domains still use a resolver that doesn’t match your intended setup, check the client and system DNS settings. Verify internal domains as your organization requires; don’t use public DNS test sites for them.
- Keep a rollback option. Save the original settings and test candidate routes outside of calls. Switching during a meeting can change your exit and interrupt the session. If new rules prevent sign-in or access to the intranet, restore the original settings first, then troubleshoot conflicts one at a time.
- ✅ Before ending the meeting, confirm that both sides can hear clearly and screen sharing works—not just that you can join.
- ✅ Make the rules your daily defaults only after colleagues can see document edits and the test repository completes a pull.
- ❌ Don’t use node latency rankings as a substitute for testing the apps, or add internal company domains to public DNS rules.