Big-endian and little-endian got their names from an argument over eggs
Danny Cohen used Swift’s satire in a 1980 networking memo that asked engineers to agree on an order and stop fighting.
The names big-endian and little-endian carry an egg argument into computer engineering. Danny Cohen borrowed the camps from Jonathan Swift for a memo about the order in which information travels. A disagreement over computer words had found a literary breakfast, with everyone still determined to crack it their own way.
“On Holy Wars and a Plea for Peace” is Internet Experiment Note 137, dated April 1, 1980, under Cohen’s name at USC/ISI. Its opening announces, “This is an attempt to stop a war.” The date sits on the document itself; the technical discussion has considerably more substance than an April Fools’ punchline.
The memo works through the ordering of bits, bytes and words during serialization. A sender’s large unit of information has to become smaller units crossing a communication medium. Once the receiver reconstructs that information, both ends need compatible assumptions about where its significant pieces belong. Agreeing about the value is insufficient if the pieces arrive with conflicting directions.
Cohen links his names to the big and little ends of a word. His appendix explains Swift’s fictional dispute over egg-breaking, complete with opposing camps and political consequences wildly out of proportion to breakfast. The joke gives a memorable name to the two technical preferences without making either group’s certainty look particularly dignified.
The practical distinction can be shown without picking a processor. A two-byte value written as hexadecimal 0x1234 can be stored with byte 0x12 first or byte 0x34 first. Either arrangement can represent the same intended value when both sides use the same rule. Reading the bytes under the opposite rule produces a different value.
That small example is worth putting into a file-format or protocol test. A value whose two bytes happen to match cannot expose the reversal, just as identical socks cannot tell you which drawer was swapped. Choosing distinct bytes makes the ordering assumption visible in the actual output instead of leaving it hidden in a function name.
The memo’s discussion is broader than the modern shorthand of two byte orders. It considers consistent ordering across multiple levels and the additional agreement needed when those levels are handled inconsistently. Compressing the whole argument into a claim that Cohen simply chose a winner would lose the reason he wrote it.
A team defining a binary format can take a modest action from that history: specify the byte order at the boundary, then show an encoded example with its expected decoded value. The internal representation can remain an implementation choice so long as the shared format is explicit. Nobody receiving the file should need access to the author’s preferred machine to guess the answer.
Cohen’s plea favors agreement over continuing the argument about the best choice. His appendix even notes a difference from Swift’s satire: separate egg eaters can choose for themselves, whereas communicating systems need a shared convention. The unresolved negotiation belongs at the interface, before two entirely reasonable implementations begin disagreeing byte by byte.
Breakfast has entered the protocol.