TL;DR
This paper presents an analytical model to quantify the performance impact of using Secure Sockets Layer (SSL) on web servers. By modeling client-server interactions and using measured data, the authors demonstrate that SSL introduces a performance penalty, especially with increased security levels, but the overhead is manageable for many applications.
What Problem Does This Paper Address?
The internet's growth has made secure communication a critical requirement for businesses and consumers. Protocols like SSL are essential for protecting data from threats like eavesdropping and modification. However, implementing these security measures requires additional computational resources on both clients and servers, potentially leading to increased response times and impacting overall system performance. While the need for security is clear, the precise performance cost of SSL, especially in terms of how it scales with user load and security strength, was not well understood. This paper aims to fill that gap by providing a quantitative analysis of SSL's performance impact on web servers.
Methods
The authors developed a closed queueing network model to represent a web server serving users who make requests. Their model distinguishes between two types of requests: "type 1" requiring secure communication via SSL, and "type 2" being normal HTTP requests. The model explicitly captures the detailed client-server interactions of a typical SSL session, including the handshake protocol and subsequent data transfer.
To parameterize their model, they conducted measurements on an existing SSL implementation (SSLeay v2). This involved instrumenting the server to precisely measure the time taken for various stages of both SSL and normal HTTP requests. The model then uses these measured service times, along with user think times and the probability of a request being secure, to analytically derive the mean response time for different scenarios. They also investigated the impact of varying the number of users, the proportion of secure requests, cryptographic algorithm strength (RSA key size and symmetric encryption algorithms), and HTML file size.
Datasets
The paper does not use pre-existing datasets for training or evaluation. Instead, the authors generated their own measurement data by running experiments on a web server built with the SSLeay [12] library. They specifically measured the service times for various stages of SSL and normal HTTP requests on a Sun Ultra 10 system. These measured values were then used as input parameters for their analytical model.
Key Insights and Ideas
- SSL incurs a tangible performance penalty: The core finding is that enabling SSL on a web server demonstrably increases response times compared to serving normal HTTP requests. This penalty is not negligible, especially as the number of users and the proportion of secure requests increase.
- Security strength directly impacts performance: The paper highlights a clear trade-off between security and performance. Increasing the strength of cryptographic operations, whether through longer RSA keys or more robust symmetric encryption algorithms, leads to a significant increase in response time.
- Public-key cryptography is a major bottleneck: The decryption of the client master key using public-key cryptography (RSA) was identified as a particularly time-consuming operation, contributing substantially to the SSL overhead.
- Symmetric encryption also adds overhead, but less so for small files: While symmetric encryption algorithms also add to the response time, their impact is less pronounced than public-key operations for smaller file sizes. However, for larger files, the overhead of symmetric encryption becomes more comparable to that of public-key cryptography.
- The model is generalizable: The analytical results are valid for a wide range of user think time and service time distributions, making the model applicable to various real-world scenarios. The approach can also be extended to analyze other secure communication protocols like SET.
Key Results
- Performance Degradation with Increased
p: For a given number of users, increasingp(the probability of a request being secure) leads to a higher mean response time. For example, with 100 users, the mean response time increases significantly aspgoes from 0.1 to 0.3. - Impact of RSA Key Size: Increasing the RSA key size from 512 bits to 2048 bits dramatically increases the processing time for public-key cryptography, from 11 msec to 481 msec. This translates to a substantial rise in mean response time for secure requests.
- Impact of Symmetric Encryption: Using stronger symmetric encryption algorithms like IDEA and Triple DES (compared to RC4) increases the mean response time. For instance, with 200 users, Triple DES results in a mean response time of around 400 msec, while RC4 is closer to 300 msec.
- File Size Matters: For larger HTML file sizes (1MB and 2MB), the performance penalty from secret-key encryption becomes more significant, approaching the impact seen with public-key cryptography.
Limitations and Future Work
- Specific Implementation: The measurements were based on a specific SSL implementation (SSLeay v2). While the model is general, the exact performance figures are tied to this implementation. Future work could involve analyzing newer SSL/TLS versions.
- Limited Client-Side Analysis: The model primarily focuses on the server-side performance. While client-side processing and network delays are acknowledged as components of the delay server, their detailed impact is not deeply explored.
- Simplified Network Model: The network delay is modeled as part of a generic "delay server" without explicit network topology or congestion modeling.
- Future Work: The authors suggest extending the modeling approach to other secure communication protocols like SET. They also imply that further investigation into the impact of larger file sizes and different network conditions would be valuable. The potential for hardware accelerators for cryptographic operations is also mentioned as a way to mitigate performance penalties.
Summary updated