Abstract
An enterprise email filtering system is introduced integrating present business systems with service plugins, which connects the email and related business parts as office automation, cloud storage etc.. The user can fulfill his business-related process in the business system with personalized rules. The system also has adopted a special white list rule for a special business, which, on basis of the traditional email filtering functions, can recognize and process important emails for users. The design has promoted user friendliness of the mailbox and the business efficiency. The system is deployed in Guangdong Province Key Laboratory of Computer Network for email service and shows excellent performance in service.
Introduction
In modern enterprise environments, email serves as a primary vehicle for business data. However, traditional email filters often operate as "black boxes," merely distinguishing between spam and legitimate mail. This creates a significant manual burden for employees who must download attachments—such as reports or assignments—and manually upload them into siloed business systems or cloud storage.
The researchers identify a significant gap: while 40.7% of Chinese enterprises had deployed information systems by 2015, these systems often remained siloed from email services. The "MATRIX" system addresses this by creating an intelligent filter that understands the business value of an email. By integrating directly with business workflows through service plugins, it automatically routes data to the appropriate destination, bridging the divide between email services and enterprise digital infrastructure.
System Architecture and Methods
The MATRIX Intelligent Enterprise Mail Filter utilizes a modular, plugin-based architecture designed for high availability and flexibility.
Network Deployment
The system is deployed using DNS MX (Mail Exchanger) record redirection. This allows the MATRIX filter to intercept all incoming mail before it reaches the final mail server, ensuring compatibility with any existing mail server software without requiring complex reconfigurations.
Modular Design
To maintain responsiveness under load, the system decouples its core functions into two primary nodes:
- Receive Node: Manages SMTP request processing, concurrency control, and frequency control.
- Process Node: Handles the heavy-duty filtering tasks, including SPF detection, blacklist checks, virus scanning, and spam detection.
Service Plugins and Classification
A key innovation is the use of service plugins that connect the filter to external business systems, such as the "Weblib" cloud disk. These plugins can be "hot-swapped"—enabled or updated without restarting the service.
The system employs a three-level classification strategy:
- Important: Mail from white-listed enterprise or personal contacts.
- Ordinary: Standard legitimate mail.
- Spam: Unsolicited or malicious mail.
Users can define personalized rules to automate workflows, such as automatically uploading attachments with specific subject lines to designated cloud storage folders.
Performance Evaluation
The system was evaluated at the Guangdong Key Laboratory of Computer Network, serving approximately 80 members. Performance testing focused on concurrency and throughput using the SMTP-Source tool.
Key findings include:
- Responsiveness: With concurrency levels below 800, the average response time remained under 1 second, providing a seamless user experience.
- Processing Latency: Total delay (including queuing and processing) stayed below 10 seconds for concurrency up to 800.
- Optimization: The researchers identified CPU-intensive modules like SpamAssassin as primary bottlenecks. They mitigated this by binding specific CPU cores to filtering processes to improve parallel processing efficiency.
Conclusion and Future Work
The MATRIX system successfully automates the transition from email communication to business data processing. While highly effective in laboratory and small enterprise settings, future work will focus on optimizing the computational cost of high-intensity filtering rules and exploring performance at larger scales. Additionally, the system's efficiency is currently linked to the performance of the destination business systems, a dependency that warrants further investigation for optimization.
Summary updated