Network Separation and Network Interconnection
1. Overview
Network Separation is a network security architecture that physically or logically divides the internal business network from the internet-connected external network so that a compromise originating outside cannot propagate into internal business systems and critical data.
Network separation arose from the recognition that a single perimeter firewall is not enough to block external threats.
When an internet-exposed endpoint is infected by malware or has its account stolen through phishing, internal business systems and databases connected to the same network also come within the attacker's reach.
National infrastructure, public institutions, and financial firms in particular — where the social impact of a breach is large — set a strong premise of "never mixing the external and internal networks at all" to reduce risk at its source.
In Korea, the direct trigger for network separation becoming an institutional requirement was a series of major breaches that began in the late 2000s.
The 2011 paralysis of the Nonghyup banking network and the March 20, 2013 cyber terror attack (which simultaneously crippled broadcasting and financial firm networks) shared a common thread: malware that entered through the internet network penetrated into the internal business network.
Accordingly, the financial sector embedded a network separation mandate in the "Regulation on Supervision of Electronic Financial Transactions," and public and national agencies did so in the "National Information Security Basic Guideline," making network separation a symbol of Korean-style security control.
A. Background and Necessity
First, the limits of perimeter security became clear.
Firewalls and intrusion prevention systems (IPS) block known attacks, but threats arriving through files downloaded from the internet or email attachments look like normal traffic and are hard to detect.
Second, damage spread through lateral movement was a problem.
Once a single internet-use endpoint is infected, the attacker uses it as a bridgehead to move to internal servers; if the internal and external networks are physically separated, this path itself is cut off.
Third, the demand for regulatory compliance was strong.
The financial and public domains handle large volumes of personal and financial transaction information, so legal liability and loss of social trust are severe when an incident occurs.
Therefore, structural and preemptive control that "eliminates the incident path altogether" was preferred over a reactive approach that "detects and responds to incidents," and network separation became its core means.
2. Types and Architecture of Network Separation
Depending on "what is separated and how," network separation is broadly divided into physical and logical separation, and a network interconnection system is combined between the two separated networks to exchange only the strictly necessary data through a controlled path.
The structure diagram below shows the classification scheme of network separation and the position of network interconnection as a whole.
flowchart TD
NS["Network Separation"] --> PHY["Physical Separation"]
NS --> LOG["Logical Separation"]
PHY --> PHY1["Business-network-only endpoint"]
PHY --> PHY2["Internet-network-only endpoint"]
LOG --> CBC["CBC (client-based virtualization)"]
LOG --> SBC["SBC (server-based virtualization/VDI)"]
PHY1 --- GW["Interconnection system (gateway)"]
PHY2 --- GW
CBC --- GW
SBC --- GW
GW --> CTRL["Transfer control·DLP·malware scan·disarm (CDR)"]
CTRL --> AUDIT["Transfer history·audit·approval"]
A. Physical Network Separation
Physical network separation completely divides the business network and the internet network into separate network equipment and separate endpoints.
Users either keep one business PC and one internet PC separately, or place two network cards and a switch in a single PC so that it connects to only one network at a time.
The greatest advantage of this approach is that, because there is no physical connection between the two networks, no path structurally exists for an internet-borne compromise to propagate into the internal network.
Because the security level is the highest, it is adopted in domains demanding extremely high security, such as defense, the national information and communications network, and nuclear/power control networks.
However, since endpoints and network lines must be built in duplicate, deployment and operating costs are high, and productivity suffers because users must switch back and forth between two environments.
For example, to reflect material gathered from the internet into a business document, one must go through a separate interconnection procedure or storage-media import control, so it does not end with a simple copy and paste.
B. Logical Network Separation
Logical network separation uses virtualization technology to divide the business area and the internet area in software within a single physical endpoint or server.
It is typically divided into the CBC (Client Based Computing) method and the SBC (Server Based Computing) method.
The CBC method separates the business area and the internet area within the user's PC through a virtualization layer, and isolates the two areas so that neither can access the other's file system and network.
Because it uses the client's CPU and memory resources, the server investment burden is small, but if the endpoint itself is infected, it can be vulnerable to attacks that bypass the virtualization layer, so endpoint security management is important.
The SBC method places a virtual desktop (VDI) or virtual business area on a central server, and the user's endpoint receives and operates only the screen information.
Because business data does not remain on the endpoint and is processed only on the server, it is advantageous for leakage control and central management, but the drawback is that server and network bandwidth investment is large and a server failure affects all users.
Logical separation costs less and offers higher user convenience than physical separation, but because a single physical resource is shared, a virtualization vulnerability or configuration error can collapse the separation boundary, so its security level is rated relatively lower.
| Category | Physical separation | Logical separation (CBC) | Logical separation (SBC) |
|---|---|---|---|
| Separation level | Full separation (highest) | Virtual separation in endpoint | Virtual separation in server |
| Resource use | Endpoint·line duplication | Client resources | Server resources·bandwidth |
| Data residue | Exists on each endpoint | Exists on endpoint | Only on server |
| Cost | High | Medium | High (server investment) |
| Convenience | Low | Medium | High (access anywhere) |
| Main application | Defense·control networks | General enterprise business network | Remote·telework |
3. Network Interconnection Technology and Data Transfer
Separating networks strengthens security, but situations arise where data must be exchanged between the internal and external networks for business purposes.
For instance, it is necessary to forward a civil complaint received at a public service portal (external network) to the processing system of the internal business network, or to export statistics generated internally to a public web server.
The technology that transfers only the strictly necessary data through a controlled path while maintaining the security effect of separation is network interconnection.
The following figure shows the data transfer flow through the network interconnection system in detail.
flowchart LR
U["Internal-network user/system"] --> IS["Internal relay server (Send)"]
IS --> ENG["Transfer engine·encryption·integrity check"]
ENG --> OWG["One-way transfer device (Data Diode)"]
OWG --> ES["External relay server (Receive)"]
ES --> CDR["Malware scan·content disarm (CDR)"]
CDR --> EXT["External network·public service"]
EXT -.->|reverse request| PX["Approval·filtering proxy"]
PX -.->|after policy check| ENG
ES --> LOG["Transfer history·audit log"]
A. Interconnection Components
A network interconnection system generally consists of an internal relay server placed on the internal network, an external relay server placed on the external network, and a transfer medium between them.
The internal relay server collects, verifies, and encrypts the data to be transferred and passes it to the transfer medium, and the external relay server decrypts and inspects the received data before delivering it to the destination system.
The key point is that the two relay servers do not establish a direct session; data flows only through the dedicated transfer medium between them (shared storage, serial transfer, data diode, etc.).
This structurally blocks arbitrary access or reverse control from the outside into the inside.
B. One-Way Transfer and Two-Way Interconnection
The most powerful form is one-way (unidirectional) transfer using a Data Diode.
Using a physical structure that allows an optical signal to flow in only one direction, transfer from the internal network to the external network is possible, but the signal path returning from outside to inside does not physically exist at all.
This is used in environments where blocking intrusion is absolutely more important than preventing leakage, such as when exporting data from a control system (ICS/SCADA) to an external monitoring network.
On the other hand, when two-way data exchange is needed, as in civil complaint processing, the request and response are each configured as independent one-way channels, and the risk is controlled by passing only data in the reverse direction that satisfies predefined protocols, formats, and approval policies.
C. Transfer Control and Content Disarm
The security value of network interconnection lies not merely in "narrowing the passage" but in making the passing data itself safe.
Therefore, during transfer, malware scanning, allowed file-type and extension restrictions, data loss prevention (DLP) policy application, and content disarm and reconstruction (CDR) are performed.
CDR removes risky elements such as macros, scripts, and executable objects from document files and reconstructs the file using only safe elements, so it has the effect of blocking even unknown malware (zero-day).
All transfer history is recorded — who sent what, when, and where — and is used for audit trails and incident investigation.
4. Comparison and Application Cases
The choice between physical and logical separation comes down to a trade-off between "security level" and "business efficiency and cost."
Physical separation offers the strongest security because it structurally eliminates the compromise path, but in an environment where telework, cloud, and generative AI use are growing, it severely constrains business productivity.
In fact, as telework became common after COVID-19, criticism grew that physical separation — which prevents access to the internal business network from outside — is an obstacle to digital transformation.
A representative public-sector case is the e-government structure that separates the government administrative network from the internet network and forwards data from the civil complaint portal to internal administrative systems via a network interconnection system.
The 2023 local administrative network outage showed that interconnection and authentication components in a separated environment can become a single point of failure, suggesting that availability design must accompany separation, not just separation itself.
In the financial domain, Article 15 of the "Regulation on Supervision of Electronic Financial Transactions" has required the internal network of the computing room to be physically separated from the internet, and banks and securities firms have operated business and internet endpoints separately.
In the defense and intelligence fields, a structure using data-diode-based one-way transfer is used to maintain the secrecy of operational and intelligence networks while still exporting necessary information externally.
Thus, even the same "network separation" varies in separation strength and interconnection method according to an organization's risk level and business characteristics.
5. Deep Dive: Paradigm Shift in Separation Regulation (N2SF·Financial-Sector Roadmap)
Recent network separation policy is undergoing a major shift from "uniform, across-the-board separation" to "differentiated security based on data importance."
After announcing the Multi-Level Security (MLS) roadmap in 2024, Korea's National Intelligence Service renamed it the National Network Security Framework (N2SF) to avoid confusion with U.S. Department of Defense terminology and released guideline v1.0 in 2025.
The core of N2SF is to classify business information and systems into three grades — Classified, Sensitive, and Open (C·S·O) — and apply differentiated security controls by grade.
That is, instead of separating all work equally from the internet, it allows the use of generative AI, cloud, and SaaS for Open-grade work while maintaining strong separation for Classified-grade work.
However, this is not "abolition of network separation" but an improvement adapted to reality; because the clause allowing communication between grades to be blocked and separated is retained, the prevailing interpretation is that separation itself has not disappeared.
The flow of replacing the uniform network separation clause with the N2SF clause through revision of the National Cybersecurity Basic Guideline is continuing, and the 2026 cybersecurity evaluation newly reflects bonus points for N2SF adoption.
In the financial domain as well, in August 2024 the Financial Services Commission announced the "Roadmap to Improve Network Separation in the Financial Sector," presenting a direction to gradually improve outdated regulations more than a decade after their introduction.
The roadmap includes converting generative AI and SaaS special exceptions validated through the regulatory sandbox into regular institutions, permitting logical separation between the research and development network and the business network, and overhauling supervisory regulations related to personal credit information processing and the cloud.
These changes show that the center of gravity is shifting from formal physical separation toward substantive security controls centered on "security controls as an alternative to network separation."
Ultimately, this flow converges with the Zero Trust paradigm — removing implicit trust based on network location and continuously verifying access according to identity, device posture, and data grade.
6. Considerations and Implications
First, it must be made clear that network separation is a means, not an end.
Separation is only a control that reduces compromise paths; overconfidence that separation alone makes one safe overlooks residual risks such as storage-media import, insider threats, and configuration errors in the interconnection segment.
Therefore, network separation should be designed as one axis of defense in depth, combined with endpoint security, access control (NAC), data encryption, and log auditing.
Second, the trade-off between security and productivity should be tuned to the organization's risk level.
Forcing physical separation on all work makes digital transformation and AI use difficult, so an approach that differentiates separation strength according to data importance — like N2SF's C·S·O grading — is realistic.
Third, the availability and single points of failure of the interconnection segment must be managed together.
If the relay server or authentication server stops, civil-complaint and external services can be interrupted, so duplication, failover, and capacity planning should be reflected at the design stage and disaster recovery drills conducted regularly.
Fourth, the transition to Zero Trust must be strategically prepared.
In an environment where cloud, telework, and partner interconnection are increasing, identity-based continuous verification and microsegmentation may be more effective than physical boundaries, so a roadmap to evolve network separation into a Zero Trust architecture should be established.
Fifth, governance response to regulatory change is needed.
When institutions change — as with N2SF and the financial-sector roadmap — organizations must jointly adjust their data classification scheme, security control redesign, and solution investment plans, and proactively prepare change management and budget.
References
- Dream Security, "National Network Security Framework (N2SF) Complete Guide": https://www.dreamsecurity.com/pr/news/1112
- ZDNet Korea, "Revised after 18 years… New National Network Security Framework unveiled": https://zdnet.co.kr/view/?no=20250123170744
- Boannews, "N2SF to roll out in full this year": https://m.boannews.com/html/detail.html?idx=142634
- Shin & Kim, "Roadmap to Improve Network Separation in the Financial Sector": https://shinkim.com/kor/media/newsletter/2533
- NIST, "Zero Trust Architecture, SP 800-207": https://csrc.nist.gov/pubs/sp/800/207/final
In one line: Network separation is a Korean-style security architecture that physically and logically divides the internal business network from the external internet to structurally block compromise propagation while transferring only strictly necessary data through controlled interconnection, data diodes, and CDR, and it is now evolving from uniform separation toward data-importance-based differentiated security and Zero Trust via N2SF's C·S·O grading and the financial-sector improvement roadmap.