| Reference | Octets (octet) | Petaoctets (Po) |
|---|---|---|
| A plain text message (160 characters) | 160 octet | 1.6 × 10-13 Po |
| A three-minute MP3 | 3000000 octet | 0.000000003 Po |
| A smartphone photo | 4000000 octet | 0.000000004 Po |
| A high-definition film | 4 × 109 octet | 0.000004 Po |
| A dual-layer Blu-ray disc | 5 × 1010 octet | 0.00005 Po |
The octet is a unit of digital information equal to exactly eight bits. Its symbol is octet, and its name comes from the Latin for a group of eight. It is the unit in which nearly all stored data is counted, and the reason it is called an octet rather than a byte is worth knowing.
A byte was originally the number of bits a machine used to hold one character, and that number was not fixed. Early computers used six-bit bytes, some used seven, and a few used nine. The IBM System/360 of 1964 settled on eight and its commercial success made eight the norm, but the word byte kept its older, machine-dependent meaning in specifications where ambiguity was dangerous.
The networking world therefore adopted octet. When a protocol standard says a header field is four octets long, no reader anywhere can interpret that as anything but thirty-two bits. The internet standards published by the IETF use octet throughout for exactly that reason, and French, Romanian and several other languages adopted the word for ordinary use as well, so that a hard disc sold in France is measured in gigaoctets.
Eight bits hold 256 distinct values, which is why so many things come in units of 256. A colour channel runs from 0 to 255. An IPv4 address is four octets, written as four numbers each between 0 and 255. The original ASCII character set used seven bits, leaving one spare, and the various eight-bit extensions that followed each filled that spare bit with a different set of accented letters — the mess that Unicode eventually replaced.
Modern text encoding still counts in octets. UTF-8 uses one octet for the unaccented Latin alphabet, two for most European accented letters and Greek and Cyrillic, three for Chinese, Japanese and Korean, and four for the rest. That is why the same sentence occupies different amounts of storage in different languages, and why a text-length limit measured in octets is not a limit on characters.
Storage capacity, file sizes, memory and network payloads are all quoted in octets or their multiples. Transmission rates are the exception, being quoted in bits per second, and the factor of eight between the two conventions is the single most common source of confusion in the field.
One octet equals 8 bits, 0.008 kilobits, or 0.001 kilooctets.
The petaoctet is a unit of digital information equal to a thousand teraoctets, or a million gigaoctets. Its symbol is Po. It is the unit of institutional storage: the scale at which data belongs to an organisation rather than to a person, and at which keeping it becomes a budget line rather than an afterthought.
A petaoctet is roughly the storage of a thousand large consumer hard drives, or the text of every book ever published several times over. In video terms it is about twenty thousand hours in ultra-high definition, or two and a half years of continuous viewing. The entire catalogue of a large streaming service, held once at each quality level, comes to a few petaoctets.
Science reached this scale first. The Large Hadron Collider at CERN records tens of petaoctets a year after its trigger systems have already discarded more than 99.99 per cent of what the detectors see, and the full archive runs to several hundred petaoctets. Astronomy, genomics and climate modelling all keep archives of comparable size, and the discipline of managing them became a research field of its own.
Commercially the petaoctet describes a single data centre's storage rather than a company's total. A large organisation holds tens or hundreds of petaoctets across many sites, and the largest cloud providers hold exaoctets. At this size the practical problems are not capacity but the electricity to keep the drives spinning, the cooling, and the certainty that a fraction of the hardware is failing at any moment.
That last point drives the design. In a petaoctet array, drive failures are not exceptional events but a continuous background rate, so the system is built to lose devices constantly and rebuild without interruption. Data is stored with erasure coding across many machines, and no single copy of anything is trusted.
Reading a petaoctet is itself a problem. Even at ten gigaoctets per second, a rate few systems sustain, a full pass takes more than a day. This is why analysis at this scale is designed to move the computation to the data rather than the data to the computation, an inversion that shaped the whole field of distributed processing.
One petaoctet equals 1,000 teraoctets, 1,000,000 gigaoctets, 8 petabits, or about 0.8882 pebioctets.