Skip to main content

How Local SASE Enforcement Reduces Latency and Cloud Backhaul

Published on:
.
5 min read

Short answer: in PoP-based SASE, traffic travels to the vendor's nearest point of presence (PoP) to be inspected, then continues to its destination. That detour typically adds 20 to 300 milliseconds per round trip. Local enforcement inspects traffic on the gateway or device it already passes through, so there's no detour. Zenarmor works this way, and its inspection adds about 0.2 milliseconds.

This article explains where SASE latency comes from, which applications notice it, what local enforcement changes, and what to measure before you choose.

What is cloud backhaul?​

Backhaul means carrying traffic somewhere else to be processed before it goes where it was headed. Older networks backhauled branch traffic to the data center firewall over MPLS. PoP-based SASE moved that checkpoint from the data center to the vendor's cloud, but the pattern is the same: traffic leaves the site, goes to the checkpoint, gets inspected, and then goes on to the internet or to another site.

When the nearest PoP is close and lightly loaded, the cost is small. When it's far away or busy, every connection pays for it.

Where does the latency come from?​

Most of the delay comes from four places.

  1. Distance to the PoP. Light in optical fiber covers roughly 200 km per millisecond. A PoP 1,000 km away adds about 10 ms of round trip before anything is inspected, and real fiber routes are rarely straight lines.
  2. Hairpinning. If the destination is close to the user but the PoP isn't, traffic travels out to the PoP and back. A user in a regional office calling a server in the same city can end up routed through another country.
  3. The trip back out. After inspection, traffic leaves the PoP for its destination. Whether that's fast depends on how the vendor's network connects to the SaaS application or website, which you can't see or change.
  4. Inspection and queuing. The inspection itself takes time, and a shared PoP serving many customers can queue traffic at busy hours.

Distance and hairpinning usually dominate. That's why PoP-based architectures show a wide range, typically 20 to 300 ms per round trip, depending on where the user is relative to the PoP.

Which applications notice it?​

A web page that loads in two seconds won't feel 30 ms slower. Some traffic is much more sensitive:

  • Voice and video calls: delay and jitter show up as talk-over and frozen frames.
  • Remote desktop and VDI: every keystroke and mouse move makes a round trip.
  • Chatty applications: anything that makes many small requests in sequence (file shares, some line-of-business apps, database clients) multiplies the delay by the number of round trips.
  • Branch to branch traffic: two offices in the same region can end up talking through a PoP elsewhere.

If your users mostly browse and use email, PoP latency may never be a complaint. If they spend the day on calls, in remote sessions, or on chatty internal apps, it usually is.

What does local enforcement change?​

With local enforcement, the inspection engine runs where traffic already flows:

  • On the gateway or firewall at each office or branch, for everything behind it
  • On a server, virtual machine, or cloud instance, for workloads where they run
  • On the endpoint, for laptops wherever they connect

Traffic goes straight from the user to its destination, and the checkpoint is on the path instead of beside it. The added delay is the inspection time on that device, not a trip across the country.

Local enforcement also changes two things besides latency:

  • Where traffic goes. Traffic doesn't pass through a vendor's cloud to be inspected. That matters to organizations with data residency or privacy requirements.
  • What happens when the cloud has trouble. If enforcement runs on the device, it doesn't depend on reaching a PoP. It's still worth asking what does depend on the vendor's cloud, such as management, updates, and reporting.

What doesn't it change?​

Local enforcement isn't free of trade-offs, and it's fair to state them:

  • The device needs capacity. Inspection runs on your hardware, so the gateway or server has to be sized for the traffic, especially with TLS inspection turned on.
  • Something has to run at each location. That can be an existing firewall, a virtual machine, or an agent on each laptop. If a site has no equipment you control, a PoP-based design asks less of it.
  • Your internet path is still your internet path. Local enforcement removes the detour. It doesn't make a slow uplink faster.

What to measure before you choose​

Vendor latency figures are only useful if you know what they measure. Test with your own traffic:

  1. Where is traffic inspected? Locally or in a PoP? If in a PoP, which one serves each of your locations?
  2. Measure round trip with and without the service. Ping and time a few real applications from each site, for example your video platform and a branch-to-branch connection.
  3. Test at busy hours. Shared infrastructure behaves differently at 10 a.m. than at midnight.
  4. Turn on TLS inspection. Most traffic is encrypted, and inspecting it is where the processing cost is. Measure with it on.
  5. Ask how the vendor measured its figure. Per packet or per connection, on what hardware, at what load.
  6. Ask what keeps working if the vendor's cloud is unreachable. Does enforcement stop, or only management?

How Zenarmor does this​

Zenarmor enforces security locally, at the point of connection, without re-routing traffic through a vendor's cloud. One app runs on a gateway, in the cloud, or on an endpoint: on OPNsense (and pfSense CE) as a one-click plugin, on OpenWRT, FreeBSD, and Linux, bare metal or virtual, with endpoint agents for Windows, macOS, and Linux. Inspection adds about 0.2 milliseconds. To size the device it runs on, see the hardware requirements.

The engine inspects traffic at layer 7, inline, including TLS. Full TLS inspection is an add-on on Unified Secure and included in Unified Secure Plus (see plans).

If Zenconsole, the cloud management console, can't be reached, gateways and endpoints keep enforcing their last policy locally until it sends a change. Since release 2.7, peers on a Zero-Trust Network also connect directly when the coordinator or relay servers are offline, and admins can choose the region where reporting data is stored.

See deployment options for how the pieces fit together, or One security app for endpoints, gateways, and cloud for the full stack.

FAQ​

Does SASE add latency? It depends on where traffic is inspected. PoP-based SASE sends traffic to the vendor's point of presence first, which typically adds 20 to 300 milliseconds per round trip. Local enforcement inspects traffic on the gateway or device it already passes through, so the added delay is the inspection time on that device.

What is cloud backhaul in SASE? Cloud backhaul is the detour traffic takes to a vendor's cloud point of presence to be inspected before it continues to its destination. It adds distance, and often a hairpin, to every connection.

How does local enforcement reduce latency? It removes the detour. The inspection engine runs on the firewall, server, cloud instance, or endpoint that traffic already flows through, so traffic goes straight to its destination. Zenarmor's inspection adds about 0.2 milliseconds.

Which applications are most affected by SASE latency? Voice and video calls, remote desktop and VDI, applications that make many small requests in sequence, and traffic between branch offices. Ordinary web browsing rarely shows a noticeable difference.

What happens to local enforcement if the management cloud is down? With Zenarmor, gateways and endpoints keep enforcing their last policy locally until Zenconsole sends a change. Enforcement runs on each device, so it doesn't depend on reaching the cloud.