Impact of security protocol on server performance

Summary

The lecture examines how Transport Layer Security (TLS) affects web-server performance and how to balance security with throughput. It uses a numerical example to compare an unsecured site with different ways of applying TLS.

Key concepts

• TLS adds processing overhead. The handshake requires extra exchanges and cryptographic work before application data is sent. During bulk transfer, the server encrypts data and the client decrypts it; hashing is also used to check data integrity.
• Overhead depends on the algorithms and key size. The example considers different key sizes and encryption and hashing options, including RC4 and MD5.
• Throughput is limited by the bottleneck. The lecture defines server throughput as approximately:

Throughput ≈ 1 ÷ service demand at the bottleneck device

The bottleneck is the resource with the greatest service demand. Improving other resources may not increase overall throughput.

• A web transaction can be slowed by several parts of the system, including the client, network, server, application server, or database and storage.

Numerical example and strategies

• No TLS: Disk access is the bottleneck in the example. With a disk service time of 0.01 seconds, the stated maximum throughput is 100 requests per second.
• TLS across the entire site: The added handshake and cryptographic work make the server the bottleneck. The stated throughput falls to 19.3 requests per second.
• Encrypt only selected pages: The lecture argues that public, non-sensitive pages do not all need TLS. For a mixed workload—described as roughly 60% secure requests and 40% insecure requests, though the subtitles give inconsistent proportions—the example reports throughput of 31.43 requests per second. The general lesson is to protect pages that need it rather than encrypting every page unnecessarily.
• Use a cryptographic accelerator: Hardware can speed up cryptographic operations. In the example, this reduces server processing enough for disk access to become the bottleneck again. The lecture presents hardware acceleration as a way to improve performance but gives no clear final throughput figure.
• Reuse cached session state: Reusing a valid cached TLS session can avoid some handshake work. The lecture considers a scenario in which 60% of connections use cached state and recommends managing session-key lifetimes appropriately. The subtitles are unclear about the calculated throughput, but the intended takeaway is that session reuse can reduce handshake overhead.

Main lesson

TLS can substantially reduce throughput when cryptographic processing becomes the bottleneck. To manage the impact, identify the actual bottleneck, apply encryption where it is needed, and consider hardware acceleration or session reuse. The numerical results are specific to the example and its assumptions.

Speakers and sources

• Speaker: One instructor or narrator discussing TLS and a numerical web-server performance example.
• Other speakers: None identified.