Join our Newsletter — 33% off our NHI Course
Home› Guides› Software Defined Perimeter (SDP) Guide: CSA Architecture, Single…
Guide Zero Trust

Software Defined Perimeter (SDP) Guide: CSA Architecture, Single Packet Authorisation and ZTNA Mapping

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 14 November 2026 15 min read
On this page

Most remote access gateways still answer every packet from the internet before they know who sent it, which is why pre-authentication flaws in exposed gateways keep turning into breaches. The software defined perimeter (SDP) is the Cloud Security Alliance (CSA) architecture built to close that gap: a controller decides who may connect, and accepting hosts stay silent until a client has proved its identity with a single cryptographic packet. This guide explains the SDP specification as an architecture, how single packet authorisation (SPA) actually works, the six deployment models, and how those ideas map onto the zero trust network access (ZTNA) products organisations buy today. It is written for network and security architects evaluating or designing private access. For the product category itself, rollout and broker selection, see the ZTNA Guide; this page stays with the architecture underneath.

Key takeaways

  • SDP is a specification, ZTNA is a market. CSA published SDP v1.0 in April 2014 and v2.0 in March 2022; ZTNA is the product category that implements the same controller and gateway pattern, with or without SDP's packet-level hiding.
  • Three roles carry the whole design. The controller decides, the initiating host (IH) asks and the accepting host (AH) enforces; in NIST SP 800-207 terms the controller is the policy decision point and the hosts are enforcement points.
  • Single packet authorisation is what makes the perimeter dark. A default-drop firewall opens a port only for the source that sent a valid, encrypted, authenticated, non-replayed packet, so scanners see nothing to attack.
  • Mutual authentication protects the data path. After SPA, the IH and AH build a mutually authenticated channel such as mTLS or IKE, so a stolen SPA key alone is not access.
  • Ask vendors which parts they actually implement. Many ZTNA services hide applications behind outbound connectors but still expose an authenticating front door; the questions in this guide separate SDP-style darkness from marketing.

What the CSA specification defines

The CSA SDP Specification v1.0 was released on 30 April 2014 as a protocol proposal inviting discussion. Version 2.0, released in March 2022, describes the architectural components, their interactions and a basic security communications protocol, and revises the relationship to zero trust, the onboarding and access workflows and the sequence diagrams for three SDP sub-protocols. It also adds an SPA message format, covers the use of UDP and discusses alternatives. CSA states that SDP aims to hide networks and resources, prevent unauthorised access to services and enforce an identity-centric access policy model.

NIST SP 800-207 recognises the same idea. Its section 3.1.3, "ZTA Using Network Infrastructure and Software Defined Perimeters", describes an overlay network, typically at layer 7, where the policy administrator acts as the network controller and configures the network from the policy engine's decisions, and notes that these approaches are sometimes referred to as SDP approaches. See the NIST SP 800-207 Zero Trust Architecture Guide for the wider model.

Components and their zero trust equivalents

SDP componentWhat it doesNIST SP 800-207 roleTypical ZTNA product equivalent
SDP controllerAuthenticates hosts, consults identity and policy services, tells AHs whom to accept and tells IHs which AHs they may reachPolicy decision point (policy engine plus policy administrator)Cloud broker or management plane, policy console
Initiating host (IH)Presents user and device identity, sends SPA, builds mutually authenticated tunnelsSubject's device with agent (device agent/gateway model)Endpoint client or agent
Accepting host (AH)Drops everything by default, admits only IHs the controller approvedPolicy enforcement pointConnector, gateway or app-side agent
SDP gatewayAn AH that protects one or more servers without changing themEnforcement point in front of a resource or enclaveConnector placed beside private applications
Identity and PKI servicesFeed the controller: directory, SAML, MFA, certificates, device signalsPolicy information points (identity, CDM, threat intelligence)IdP integration, device posture feeds

The access workflow

The specification's workflow keeps the control plane separate from the data plane. In order:

  1. One or more controllers come online and connect to authentication and authorisation services such as PKI, SAML, LDAP and MFA.
  2. Accepting hosts come online and authenticate to the controllers. From then on they ignore all other hosts and unprovisioned requests.
  3. Each initiating host authenticates to a controller when it comes online, sending SPA first so the controller itself stays hidden.
  4. The controller decides which AHs this IH may reach, given identity, device and context.
  5. The controller instructs those AHs to accept the IH, with any encryption policy.
  6. The controller sends the IH its list of permitted AHs.
  7. The IH sends SPA to each AH and then builds a mutually authenticated connection to it.

CSA guidance recommends more than one controller for load balancing and to avoid a single point of failure, and every component can be made redundant.

How single packet authorisation works

SPA is the part of SDP that most products leave out, and the part that most changes the attack surface. The best documented implementation is the open source fwknop project, which describes SPA as a cryptographic authentication and authorisation layer in front of a default-drop firewall.

  • Default drop: The protected host's firewall drops all inbound traffic to the service. Nothing listens visibly; in fwknop's default mode the daemon sniffs packets with libpcap and a BPF filter rather than binding a socket, and even its optional UDP listener never acknowledges SPA packets.
  • The packet: One UDP datagram (fwknop's default is port 62201) carries 16 bytes of random data, a username, a timestamp, the client version, the message type and the access requested, so no two packets are identical.
  • Protection: The payload is encrypted with AES (Rijndael) or GnuPG and authenticated with an HMAC; fwknop's documentation gives SHA-256 as the default digest.
  • Replay defence: The server stores a digest of every valid packet and refuses a repeat, logging the attempt, regardless of the source address the replay arrives from.
  • Source address inside the payload: The address to be allowed can be carried inside the encrypted payload, so the network-layer source address is not trusted.
  • Short-lived rule: A valid packet opens the port only for that source, and the rule expires; the session that follows still has to authenticate on its own.

SPA compared with older hiding techniques

TechniqueHow access is requestedWeaknessesWhere it fits
Open port with loginConnect, then authenticate at the application or gatewayEvery pre-authentication flaw is reachable by anyone; service is discoverable by scanningPublic services that must be reachable
Port knockingA sequence of connection attempts to closed portsTwo bytes of data per packet, delays between packets, spoofable mid-sequence, looks like a port scan to an IDSLargely superseded
Single packet authorisationOne encrypted, authenticated, non-replayable packetNeeds a client and key distribution; does not protect against client-side compromiseControllers, gateways and admin interfaces that should be dark
Outbound-only connectorConnector dials out to a broker; no inbound listener at the siteThe broker's front door is public and becomes the exposed surfaceMost ZTNA services for private applications

SPA is not security through obscurity alone: the packet is a cryptographic credential, and the hiding is a consequence. The value of concealment is still real, because a scanner cannot list a service it cannot see. The 2024 Ivanti Connect Secure exploitation is a clear illustration of what SDP was designed to remove: chained pre-authentication flaws in an internet-facing gateway were reachable by anyone who could send it a request.

The six deployment models

SDP v2.0 defines six deployment models. In each, connections between components are protected by mutually authenticated protocols such as mTLS or IKE, and the controller is present even where diagrams omit it.

ModelShapeTypical use
Client-to-gatewayClients reach a hidden gateway that protects one or more unchanged serversRemote access to legacy and private applications, the classic ZTNA case
Client-to-serverThe service and the AH run on the same host, secured end to endServers where you can install an agent and want no gateway hop
Server-to-serverAll servers hidden in the perimeter, connections encrypted whatever the networkEast-west traffic, cloud workloads, APIs between services
Client-to-server-to-clientPeer traffic passes through a hidden intermediary serverVoice, chat and video relays
Client-to-gateway-to-clientPeer protocols that need direct client connections, with a hidden gateway enforcing policyPeer-to-peer applications under policy
Gateway-to-gatewayGateways hidden in the perimeter; new in v2.0Some IoT environments and site links

Server-to-server is where SDP overlaps with microsegmentation and workload identity. The Microsegmentation vs ZTNA guide explains when host-based segmentation is the better tool for east-west traffic.

How SDP maps to today's ZTNA products

CSA's 2020 paper on SDP and zero trust presents SDP as a way to implement zero trust networks. The ZTNA products on the market implement the controller and enforcement split almost universally, but implement SDP's other elements unevenly. Use this table to see which SDP properties a given service delivers.

SDP propertyHow to recognise it in a productQuestion to ask
Separate control and data planesA management or broker plane that never carries application traffic, or does so only as a relayCan the data path survive a broker outage for established sessions?
Authenticate before connectNo application or gateway port answers an unauthenticated clientWhat does an unauthenticated scanner see on your connector and broker addresses?
SPA or equivalentA cryptographic first packet, or an outbound-only connector with no inbound listenerIs the broker's own entry point protected by SPA, or is it a public login page?
Mutual authenticationDevice certificates or keys on the client, mTLS to the enforcement pointHow are client certificates issued, rotated and revoked when a device is lost?
Per-session, per-application authorisationPolicy names applications and ports, not subnetsCan policy grant a single port on a single host to a single group?
Dynamic revocationController can tear down live sessions when risk changesHow fast does a disabled account or failed posture check end an active session?

The outbound connector pattern deserves care. It removes inbound listeners at the application site, which delivers much of SDP's benefit, but it moves the exposed surface to the broker. That is acceptable when the broker is a hardened, provider-operated service with its own monitoring, and a risk when it is a self-hosted appliance on a public address. For the operational side, see Securing the ZTNA broker and remote access VPN itself, and for the VPN comparison see ZTNA vs VPN.

Decision guidance: when SDP-style darkness is worth it

  • Admin interfaces and jump hosts: SSH, RDP and management planes that only a small, managed population uses are the best fit for SPA. The client burden is low and the reduction in exposure is large.
  • Self-hosted controllers and gateways: If you run your own broker or gateway on a public address, hiding it behind SPA removes it from opportunistic scanning and mass exploitation.
  • Unmanaged and third-party users: SPA needs a client and a key. Where you cannot install a client, use a clientless resource portal (NIST SP 800-207's resource portal model) and accept that its front door is public; harden and monitor it accordingly.
  • East-west and workload traffic: The server-to-server model works when you control both ends; in Kubernetes and cloud estates, workload identity with mTLS often delivers the same mutual authentication with less custom tooling.
  • Small organisations: Buy a managed ZTNA service with outbound connectors and keep SPA for the few admin services you must expose; running your own SDP controller is rarely justified without staff to operate it.
  • Large or regulated organisations: Require redundant controllers, documented key and certificate lifecycle, and SPA or outbound-only exposure for every enforcement point; record the design against NIST SP 800-207 deployment models.

Implementing SDP principles

  1. Inventory exposure: list every internet-reachable remote access, admin and gateway endpoint, and what an unauthenticated scan returns from each.
  2. Choose the model per application: client-to-gateway for legacy apps, client-to-server where agents can run, server-to-server for service traffic.
  3. Establish identity sources first: IdP with phishing-resistant MFA, device certificates or device identity, and posture signals the controller can consume. See the Device Trust and Posture Guide.
  4. Design key and certificate lifecycle: how SPA keys and client certificates are issued, stored (hardware-backed where possible), rotated and revoked.
  5. Pilot in parallel: run the SDP or ZTNA path alongside existing access for a small IT group, with default-drop rules applied to a test host first.
  6. Close the old door: once the pilot is stable, set the protected service's firewall to drop all traffic except from AHs or SPA-opened sources, and confirm with an external scan.
  7. Roll back safely: keep an out-of-band break-glass route (console access or a separately protected admin path) so a failed controller or lost key does not lock administrators out; test it before closing the old door.

Where identity fits

SDP is identity-centric by design: every IH and AH has an identity, and machine-to-machine models make those identities non-human. SPA keys, client certificates and gateway credentials are secrets that need owners, rotation and revocation like any other machine credential. A shared SPA key baked into a golden image, or a connector token that never expires, quietly recreates the static trust SDP was meant to remove. Treat them under the Identity-Centric vs Network-Centric Zero Trust model and your secrets programme.

Detection and measurement

SPA produces useful signals: a dark service should see almost no traffic, so anything that reaches it is meaningful.

  • Controller logs: host authentication, policy decisions, denied requests and session teardown, sent to the SIEM.
  • SPA daemon logs: invalid HMACs, decryption failures and replay detections indicate key misuse or probing.
  • Firewall logs on AHs: drops to protected ports from sources that did not send SPA, which should be the only traffic a scanner generates.
  • Session telemetry: duration, bytes and destinations per IH, to spot a compromised client using legitimate access.

Metrics that show it is working:

  • Number of remote access and admin endpoints answering unauthenticated requests from the internet (target: zero, or a short documented list).
  • Percentage of private applications reachable only through the controller and enforcement points.
  • Median time from account disablement or posture failure to termination of active sessions.
  • Percentage of SPA keys and client certificates within their rotation period.
  • Replay and invalid-HMAC events per month, each triaged.

Common failure patterns

  • Calling an exposed login page SDP: a broker or gateway that serves a sign-in page to anyone is authenticate-after-connect, whatever the brochure says, and inherits every pre-authentication flaw.
  • One shared SPA key for everyone: a single key in every client means one leaked laptop opens the door for all; use per-user or per-device keys.
  • Single controller: without redundancy the controller becomes the outage, and teams respond by opening ports "temporarily".
  • Leaving the old path open: SDP is deployed but the VPN or direct firewall rule stays, so attackers use the old route. Unused remote access accounts on that route are a known breach pattern, as the Colonial Pipeline attack showed.
  • Subnet-wide grants: mapping a VPN group straight into an SDP policy that allows whole networks keeps lateral movement possible.
  • No break-glass: the default-drop rule is applied before an out-of-band route is tested, and a key or controller failure locks out administrators.

Framework mapping

FrameworkReferenceHow SDP applies
NIST SP 800-207Section 3.1.3; deployment models 3.2.1 device agent/gateway, 3.2.2 enclave, 3.2.3 resource portalSDP is the network infrastructure approach; client-to-gateway maps to device agent/gateway and enclave models
NIST SP 800-53 Rev 5SC-7 Boundary Protection, AC-17 Remote Access, AC-3 Access Enforcement, IA-3 Device Identification and Authentication, SC-8 Transmission Confidentiality and Integrity, SC-23 Session AuthenticityDefault-drop enforcement, controlled remote access, device identity, mutually authenticated sessions
NIST CSF 2.0PR.AA-05, PR.IR-01, PR.PS-04, DE.CM-01Least privilege access policy, protection from unauthorised logical access, logging and network monitoring
CIS Controls v8.1Control 12 Network Infrastructure Management, Control 13 Network Monitoring and DefenseSecure remote access architecture and monitoring of the access path
ISO/IEC 27001:2022 Annex AA.8.20 Networks security, A.8.22 Segregation of networks, A.8.5 Secure authenticationNetwork controls, separation of protected services, strong authentication before access
MITRE ATT&CK v20T1595.001 Scanning IP Blocks, T1046 Network Service Discovery, T1190 Exploit Public-Facing Application, T1133 External Remote ServicesSPA removes services from scanning and pre-authentication exploitation; controller policy limits abuse of remote services with valid credentials

Practitioner checklist

  • Run an unauthenticated external scan of every remote access, admin and broker address and record what answers.
  • Map each private application to an SDP deployment model and an enforcement point.
  • Require authenticate-before-connect for self-hosted gateways and admin services, using SPA or outbound-only connectors.
  • Issue per-user or per-device SPA keys and client certificates, hardware-backed where possible, with rotation and revocation.
  • Deploy at least two controllers and test failover.
  • Write policy per application and port, never per subnet.
  • Test and document a break-glass route before applying default-drop rules.
  • Remove the legacy access path and its accounts once the SDP path is live.
  • Send controller, SPA and enforcement logs to the SIEM and alert on replay and invalid-HMAC events.
  • Ask every ZTNA vendor the property questions in this guide and record the answers.

Standards and references

Related NHI Mgmt Group resources: ZTNA Guide · ZTNA vs VPN · Microsegmentation vs ZTNA · CSA CCZT Zero Trust Certificate Guide · Identity-Aware Proxy Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 14 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.