Skip to content
Notifications
Clear all

Am I the only one who misses a simple 'per-app' VPN mode instead of this always-on tunnel?

3 Posts
3 Users
0 Reactions
1 Views
(@alexm)
Reputable Member
Joined: 2 weeks ago
Posts: 160
Topic starter   [#21809]

Having recently completed a comprehensive POC of Netskope's ZTNA solution against a legacy per-app VPN, I find myself compelled to document a significant architectural divergence. The industry's wholesale shift to an always-on, user/device-tunnel model, while superior for continuous security posture validation, introduces non-trivial workflow and performance complexities that are often glossed over in vendor white papers.

My primary contention is that the elimination of a discrete, user-initiated 'per-app' connection mode creates friction in several specific scenarios:

* **Developer & Administrative Workflows:** Tasks requiring brief, high-privilege access to a single internal tool (e.g., a database admin console, a CI/CD server) now necessitate the entire device's traffic to be routed through the corporate tunnel for the duration. This forces all other personal/non-work traffic to suffer the latency of the corporate egress point, a clear inefficiency.
* **Bandwidth-Sensitive Applications:** In a per-app model, a user could route only SSH or RDP traffic for a server management session. With an always-on tunnel, a simultaneous personal video stream or large download is unnecessarily routed through the corporate infrastructure, consuming pooled bandwidth and potentially violating acceptable use policies.
* **Troubleshooting & Network Differentiation:** The binary state of the tunnel—either entirely on or off—complicates diagnosing whether a connectivity issue is with the user's local network, the ZTNA service, or the target application. A per-app mode provided an implicit segmentation for testing.

The Netskope Client, while highly configurable via its `settings.json`, appears philosophically opposed to this selective tunneling concept for ZTNA resources. Consider a simplified legacy VPN configuration for routing only specific subnets:

```json
// Hypothetical legacy per-app config (not Netskope's)
{
"ConnectionName": "Corp-App-Access",
"TunnelMode": "split",
"Routes": [
"10.10.1.0/24",
"192.168.100.0/24"
],
"DNS": "10.10.1.10"
}
```

In contrast, Netskope's model is predicated on a default route for all traffic (`0.0.0.0/0`) with exclusions, which inverts the logic. The administrative overhead then shifts to meticulously defining and maintaining **Exclude** lists for all non-corporate destinations, which is both fragile and infinite.

I understand the security rationale: the always-on tunnel enables real-time posture checks and prevents credential-based attacks on isolated apps. However, from a systems architecture perspective, this feels like a regression in granular control. The cost is paid in user experience, unnecessary network load, and increased support tickets for "slow internet" when the tunnel is active.

Has anyone else devised a pragmatic workaround within Netskope's policy framework to approximate a user-controlled, application-specific tunnel? Or are we to accept that this level of network-layer granularity is an intentional casualty in the evolution from VPNs to ZTNA? I am particularly interested in any quantifiable data on bandwidth overhead or latency impact from forcing all traffic through the service, as my initial benchmarks showed a 8-15% increase in latency for non-corporate traffic during tunnel sessions.



   
Quote
(@daisym)
Estimable Member
Joined: 2 weeks ago
Posts: 63
 

Oh, I feel this so much! You're spot on about the developer workflow friction. We had a similar griping session here when our analytics team rolled out a new always-on setup.

It completely borked their local development environments because every API call to test a microservice got routed through the corporate tunnel, adding crazy latency to their feedback loop. Suddenly, a quick build-and-test cycle took minutes, not seconds. They ended up creating isolated local network segments just to work around it, which felt like a total step backwards.

It's that lack of granular control that stings. There's a real difference between securing all traffic and letting someone quickly tunnel into one specific thing without dragging their whole internet connection with them.



   
ReplyQuote
(@baller_analytics)
Estimable Member
Joined: 2 months ago
Posts: 144
 

Forcing all personal traffic through the corporate tunnel just for a quick console check is a design failure. It trades minor convenience for real user productivity.

You see the same lazy architecture in analytics. Vendors push always-on, full-stream tracking under the banner of "complete data," ignoring how it bogs down apps and floods you with useless noise. Granular event control isn't a legacy feature, it's a requirement.

The latency hit on non-work traffic isn't just an inefficiency, it's a measurable performance tax that will show up in your employee satisfaction surveys. Has anyone actually run the numbers on that productivity loss versus the supposed security gain?


If it's not a retention curve, I don't care.


   
ReplyQuote