Concepts and theorems · Intermediate

The end-to-end principle: put responsibility where completion can be established

Choose validation boundaries from application requirements rather than assuming that reliable hops imply a completed task.

Ask Logeia about this articleTry the three questionsReview your answers

Updated

On this page
  1. The endpoints know the requirement
  2. Define a file-transfer completion condition
  3. Lower-layer support can still be valuable
  4. State security and durability requirements explicitly
  5. Ask Logeia about this articleCheck your understanding

The endpoints know the requirement

The end-to-end argument concerns where a system function belongs. When lower layers cannot fully establish an application’s correctness requirement, the endpoints that understand that requirement must also take responsibility. Detecting corruption on every network hop does not cover errors while reading the source or storing the destination. The argument does not ban useful lower-layer mechanisms. It distinguishes the place where the required end result can be established from mechanisms that improve efficiency, reduce common errors or make recovery cheaper. The right boundary follows the actual requirement, not a slogan that every function belongs in the application.

Define a file-transfer completion condition

Suppose a file transfer must leave a saved destination file matching the content the sender intended. A TCP acknowledgement does not certify that the receiving application persisted and verified that file. An endpoint checksum or hash comparison can address a wider part of the path, but the design must specify which bytes were read, whether the reference value is trustworthy and where verification occurs relative to saving. “Sent successfully” is ambiguous unless it names the completion boundary. Transport progress and application completion are different observations, even when both are useful pieces of a reliable transfer protocol.

Lower-layer support can still be valuable

Link recovery and TCP retransmission can remain valuable even when the endpoint must verify the final result. Repairing a small loss locally can be cheaper than restarting a large application task. Conversely, complex state in intermediate components can make recovery or evolution harder. Compare which layer has enough information and what recovery costs it can avoid. The decision is not a forced choice between endpoint-only and intermediate-only validation. A defensible design can combine them while explaining which mechanism establishes the requirement and which reduces the expected cost of reaching that verified end state.

State security and durability requirements explicitly

Detecting accidental corruption is different from protecting against an adversary who can alter both data and an unauthenticated reference value. The latter needs an appropriate authentication and key-management story. Completion can also mean receipt by an application, durable storage on one machine or durable replication across sites. Each requirement needs different evidence. State the user-visible condition, trusted participants and covered failure modes before applying the principle. Then examine what the lower-layer guarantees leave unverified. This makes the argument a practical method for identifying missing responsibility rather than a universal prescription for one network architecture.

Reading and practice are recorded separately in this browser, for each language.

Learn this in order

Understand the systems behind the web · STEP 10

Choose a learning path