Understanding the Core Technologies Behind Modern SASE Products

Marketing material tends to describe converged network security as a single capability. Underneath, it is an assembly of technologies built separately that now share one control plane: an edge network of cloud locations, traffic steering, inspection engines, an identity layer and a policy engine deciding what each session may do. Knowing how those parts behave helps during evaluation, because most deployment problems trace back to one component rather than the architecture as a whole.

Points of Presence: Where SASE Inspection Happens

Everything else depends on where inspection happens. Providers operate points of presence that terminate user and branch traffic, apply policy and forward the session onward. Two design details matter more than the published location count: whether each site runs the full set of security functions, and whether traffic is inspected once or chained between separate services. Published overviews of the core components of SASE products usually group them into a networking layer, a security layer and the policy engine that binds the two. Single-pass designs decrypt, inspect and re-encrypt once, keeping added latency predictable. Chained designs repeat parts of that work, and the cost shows up on latency-sensitive applications such as voice and virtual desktops.

SD-WAN and Agent-Based Traffic Steering

The networking layer decides how a session reaches the edge. At branch sites, SD-WAN devices measure loss, latency and jitter across available links, then steer application flows accordingly, keeping real-time traffic on the cleanest path while bulk transfers use whatever capacity remains. For individual users, a lightweight agent does the same job, choosing a nearby edge location and holding the tunnel as networks change between office Wi-Fi, home broadband and mobile data. Agentless access through a browser covers contractors and unmanaged devices, with narrower control over the endpoint in exchange. Underlay choice still matters, because broadband and mobile links vary in ways that private circuits did not.

TLS Inspection and the Limits of Decryption

Almost all traffic worth examining is encrypted, so the inspection layer has to decrypt it, scan it and re-encrypt before forwarding. That requires certificate trust on endpoints and careful exclusion lists for banking, healthcare and privacy-sensitive destinations.

Protocol changes have made this harder rather than easier. The current TLS 1.3 protocol specification confirms that all handshake messages after the server hello are now encrypted, removing visibility that older inspection techniques relied on. Encrypted client hello and QUIC-based transport continue the trend. The practical result is that identity signals, DNS telemetry and application fingerprints matter more as policy inputs, and that decryption capacity needs sizing against realistic peaks rather than averages.

Identity Providers and Device Posture as Policy Inputs

The policy engine evaluates who is asking, from what device and for which application. It draws on an identity provider through SAML or OIDC, device posture from an agent, group membership and context such as location or time. Authentication strength is part of that calculation, which is why phishing-resistant authentication using public key cryptography matters to the architecture: credentials that cannot be replayed from a lookalike site remove the most common route to a valid session. Weak identity data undermines everything above it, because granular policy expressed against inaccurate groups produces confident, incorrect decisions.

ZTNA Brokered Access Instead of Network Routes

Zero trust network access works differently from a VPN at a mechanical level. Private applications sit behind a connector that makes an outbound connection to the provider’s edge, so no inbound port is published to the internet. When an authorised user requests the application, the broker stitches the two sessions together. The user reaches that application only, and the network holding it is never routable from the client. It also makes acquisitions and third-party access simpler, since publishing an application does not require merging address space.

ZTNA vs VPN: Private Application Access Compared

Dimension Traditional VPN ZTNA broker
Connection direction Inbound listener published Outbound connector only
Access granted Network segment or subnet One named application
Client reach Routable path to the network No route to the host network
Policy inputs Credentials at connection time Identity, device posture, context
Third-party onboarding Address space and routing changes Publish the application alone
Session visibility Tunnel-level logs Per-application session records

Firewall-as-a-Service for Non-Web Traffic

Web and SaaS sessions dominate volume, but outbound device telemetry, mail protocols, database connections and site-to-site flows are not web traffic. Firewall-as-a-service handles those at the edge, applying port, protocol and segmentation rules that the web gateway never sees. Teams that map every control to browser activity usually discover this gap during their first incident review. Segmentation rules carry over from the data centre as well, where east-west controls once sat on internal firewalls that are no longer in the path once branches connect directly to cloud locations.

DLP and CASB Controls for Data and SaaS Visibility

The data layer uses several detection methods, each with different accuracy:

  • Pattern matching for structured identifiers such as card and account numbers
  • Document fingerprinting for known sensitive files
  • Classification labels applied by the business or an upstream tool
  • Machine-learning classifiers for unstructured content, which need tuning before enforcement

Most teams therefore begin in monitor mode well before enforcing any of these rules, using the first weeks to measure how often each method fires and how many of those hits turn out to be genuine.

CASB functions reach SaaS platforms two ways. Inline control sees traffic as it passes the edge and can block in real time. API-based control connects to the tenant itself and inspects data already stored there, including files shared before the platform was deployed.

What to Test When Evaluating SASE Products

  • Measure added latency with decryption enabled, not in bypass mode
  • Check whether policy objects support the granularity your entitlement model needs
  • Confirm log retention, export format and whether raw logs leave the platform
  • Test failure behaviour when the agent cannot reach its nearest edge location
  • Verify where decryption occurs relative to data residency commitments

Run these tests with production traffic patterns rather than synthetic samples. A proof of concept built on a handful of well-behaved web sessions rarely surfaces the behaviour that matters once legacy applications and bulk transfers are in scope.

Conclusion

The value of convergence is a shared policy and a single place to see what happened. The engineering underneath is still made of specific, testable parts, and they fail in specific ways: decryption capacity, identity accuracy, connector placement, classification tuning. Evaluations that test those parts individually beat comparisons built on platform-level feature lists.

Frequently Asked Questions

Do SASE products require an endpoint agent?

Not always. Branch traffic reaches the edge through a device or tunnel, and browser-based access covers unmanaged devices. An agent is needed for device posture signals and consistent steering on managed laptops.

Can SASE inspection work without decrypting traffic?

Partially. Destination, certificate metadata, DNS queries and traffic patterns support useful policy, though file scanning requires decryption.

How is private application access published safely with ZTNA?

Through outbound connectors beside the application, which avoids opening inbound ports or exposing the hosting network to clients.

What causes poor performance after a SASE deployment?

Distance to the nearest edge location, undersized decryption capacity, and overlapping policies that force repeated inspection of the same session.