Research Unix vs System V vs BSD is not a comparison of three products. It is a comparison of three numbering systems, maintained by three different organisations, over three different code trajectories — and collapsing them into a single version line is the most common factual error in writing about Unix. “V7” and “System V” look like neighbours in one sequence. They are not. V7 is the Seventh Edition of a Bell Labs manual, dated January 1979. System V is an AT&T commercial product line whose first release shipped in January 1983, out of a different department, sold to different customers, for different reasons.
This is not pedantry with a beard on it. The two commercial lineages made incompatible decisions about ordinary things — how you list processes, how the machine boots, whether echo interprets a backslash escape — and those decisions are still live in the shell you used this morning. ps aux and ps -ef are not two spellings of one command; they are two lineages’ option grammars, and the reason both work on Linux is that Linux refused to pick a side. Get the version schemes straight and a large amount of otherwise arbitrary Unix trivia turns into a legible org chart.
The Three Schemes, Side by Side
| Research Unix “Editions” | System V “Releases” | BSD “Distributions” | |
|---|---|---|---|
| Notation | First–Tenth Edition; “V7” = Seventh Edition | System V Release 1–4; SVR1–SVR4 | 1BSD, 2BSD, 3BSD, 4.1–4.4BSD, 4.4BSD-Lite |
| Who produced it | Computing Science Research Center, Bell Labs (Thompson, Ritchie, McIlroy et al.) | AT&T’s UNIX Support Group and its successors, ending at Unix System Laboratories | Computer Systems Research Group (CSRG), UC Berkeley |
| What a number counts | An edition of the UNIX Programmer’s Manual | A commercial product release | A tape distribution |
| Active span | 1971–1989 | 1983–1993 | 1978–1995 |
| Purpose of the work | Research; internal use at Bell Labs | Revenue; a supported, saleable product with a stable documented interface | Improving and extending a system the recipient already licensed from AT&T |
| How you got it | Cheap licence with source, mostly to universities | Commercial source licence, five and later six figures | Free or near-free tape — but only legal if you already held an AT&T source licence |
| The one release to know | Seventh Edition (V7), January 1979 | SVR4, announced October 1988 | 4.4BSD-Lite, June 1994 |
Three schemes, three organisations, three definitions of what a version is. The Research group numbered documentation, because its output was knowledge. AT&T’s commercial arm numbered products, because its output was revenue. Berkeley numbered tapes, because its output was a delta on somebody else’s licensed system. Every downstream confusion starts with assuming those three activities share a counter.
Research Unix: an Edition Is a Manual Revision
Research Unix came out of the Computing Science Research Center at Bell Labs — the group that Thompson and Ritchie worked in, and the one that later abandoned Unix for Plan 9. Crucially, its versions were never product releases. They were revisions of the UNIX Programmer’s Manual, and the system got named after the book. “Version 7” is shorthand for “the system described by the seventh edition of the manual,” which is why the record’s dates are manual dates and why the numbering is loose at the edges: an Edition is a documentation snapshot of a continuously modified research system, not a shipped artifact with a release engineering process behind it.
| Edition | Manual date | What it is remembered for |
|---|---|---|
| First | November 1971 | The PDP-11 system, in assembly. Not written in C — C did not exist yet |
| Second | June 1972 | Incremental; the C language is emerging alongside it |
| Third | February 1973 | Pipes ship as a primitive (Ritchie’s own paper says 1972; the manual record says 1973) |
| Fourth | November 1973 | The kernel rewritten in C — the decision that bought portability, at roughly one-third more memory |
| Fifth | June 1974 | Contemporary with the 1974 CACM paper that introduced Unix to the wider computing world |
| Sixth | May 1975 | First edition widely distributed outside Bell Labs; the kernel Lions annotated line by line |
| Seventh | January 1979 | The common ancestor. Bourne shell, awk, tar, and the first appearance of ioctl |
| Eighth | February 1985 | VAX-based, internal; absorbs 4.1BSD work |
| Ninth | 1986 | Internal; the research group is already moving toward Plan 9 |
| Tenth | 1989 | The last. Its manual was published commercially in 1990 and is readable online |
The Seventh Edition is the pivot, and the reason both other schemes exist downstream of it. AT&T’s commercial line grew out of the same code base through the support-group branch, and Berkeley’s first complete system was built on 32V, the VAX port of the Seventh Edition. If you need one sentence: nearly everything after 1979 descends from V7, and V7 is an Edition, not a release.
System V: a Different Department, a Different Business
The commercial line was never the research group’s work. It came from AT&T’s UNIX Support Group, which became the UNIX System Development Laboratory and eventually Unix System Laboratories — a support-and-productisation organisation with customers, contracts, and a mandate the research group did not have. That organisational split is the whole reason two numbering schemes ran in parallel for a decade.
The timing is routinely botched. AT&T’s freedom to run a computer business arrived with the Modification of Final Judgment approved on 24 August 1982, not with the divestiture that took effect on 1 January 1984. System III shipped externally in 1982 and System V Release 1 in January 1983 — a year before the breakup people usually credit. “After the 1984 breakup, AT&T started selling Unix” is wrong twice over: the date is off, and AT&T had been licensing Unix commercially for years already, at a Sixth Edition price commonly reported at around US$20,000 by the late 1970s. What the 1956 consent decree had prevented was a supported product line, captured in AT&T’s own stated policy: no advertising, no support, no bug fixes, payment in advance.
Two more things about the notation. First, the “V” is a Roman five, not an abbreviation of “version” — so SVR4 carries two numbers: System V, Release 4. Second, System V is where AT&T began treating the interface as a document rather than a tape: the System V Interface Definition (SVID), published in 1985, came with a verification suite, and that “conform to the spec, not to my source” logic is the direct ancestor of POSIX, the Single UNIX Specification, and today’s certification programme. SVR3 added STREAMS, the TLI networking API, and shared libraries. Then SVR4 tried to end the fragmentation it had helped create.
BSD: Add-Ons That Progressively Ate Their Host
Berkeley’s line began as supplements, and this is the part most summaries get backwards. 1BSD (9 March 1978) was not an operating system. It was an add-on tape for Sixth Edition Unix — a Pascal compiler and the ex editor. 2BSD (May 1979) was the same idea grown larger, still an enhancement package for PDP-11 sites, and the vehicle by which vi and the C shell reached the world. Only with 3BSD (end of 1979), built on the 32V VAX port, did Berkeley ship something you could boot, with demand-paged virtual memory attached.
From there the CSRG replaced more and more of what it had started by decorating: 4.1BSD (1981), then 4.2BSD (1983) — TCP/IP, the socket API, and the Fast File System, developed under DARPA contract — then 4.3BSD (June 1986). Berkeley was, by then, where most of the interesting engineering happened, but a BSD tape was still legally worthless to anyone without an AT&T source licence. Berkeley’s improvements sat on top of code it did not own.
That dependency is what produced the endgame. CSRG stripped AT&T code to ship Net/1 (June 1989) and the much larger Net/2 (June 1991), which was enough of a system that the Jolitzes built the bootable 386BSD on it in 1992. Unix System Laboratories sued Berkeley Software Design that year; the preliminary injunction was denied, and the case settled in February 1994 — three files removed of the 18,000 in Net/2, roughly seventy modified to carry USL copyright notices. 4.4BSD-Lite (June 1994) is the result, and every BSD alive today rebased onto it. A BSD number, in other words, counts tapes: what Berkeley put on a reel and posted, first as an accessory, finally as a system that had eaten its host.
SVR4: the Merge, and the Conventions War It Set Off
System V Release 4 was explicitly a merge. AT&T and Sun built it jointly, unifying System V with 4.3BSD, SunOS, and Xenix: TCP/IP, sockets, UFS, and the C shell came in from BSD; the virtual filesystem interface, NFS, the memory-mapped virtual memory system, and the shared library model came from SunOS; x86 device drivers and Xenix binary compatibility came from Xenix. It was announced on 18 October 1988 and reached customers inside commercial products from early 1989 — which is exactly why you will see both years cited, and why it is worth attributing rather than asserting.
The merge did not calm the market; it triggered the Unix wars’ second phase. Other vendors read an AT&T–Sun collaboration as capture of the standard’s direction, met at DEC’s Palo Alto offices in January 1988, and announced the Open Software Foundation in May 1988; AT&T’s counter-bloc, Unix International, followed later that year. What the two camps were fighting over was largely whose conventions would win — and because neither fully did, the split fossilised into the command surface instead of being resolved. SVR4 itself shipped the compromise as an archaeological layer: /usr/ucb, a directory holding the BSD versions of commands that had a System V version elsewhere on the same machine.
| Concern | System V way | BSD way | Where it ended up |
|---|---|---|---|
| Listing processes | ps -ef — options grouped and dash-prefixed | ps aux — options grouped and not dash-prefixed | Linux procps accepts both plus GNU long options; on FreeBSD, -e still means “display the environment“ |
| Init and boot | /etc/inittab, numbered runlevels, rcN.d symlink farms with S/K prefixes | /etc/rc and /etc/rc.local, later /etc/rc.conf plus a dependency-ordered rc.d | systemd on mainstream Linux; launchd on macOS; rc.d on FreeBSD |
| Printing | lp, lpstat, cancel | lpr, lpq, lprm | CUPS ships both command sets to this day |
| Terminal I/O | termio, then termios | sgtty | POSIX standardised termios |
| Signals | Handler reset to default on delivery; interrupted syscalls fail | Reliable signals; restartable syscalls (sigvec) | POSIX sigaction — BSD semantics under a System V-style API |
| Networking API | STREAMS with TLI/XTI | Sockets | Sockets, universally. STREAMS is a museum piece |
| Filesystem | s5fs | FFS, and its UFS descendants | Nobody ships s5fs |
echo | Interprets \n, \c inside its arguments; no -n | -n suppresses the newline; no escape interpretation | POSIX declares the case implementation-defined and tells you to use printf |
That last row is the cleanest miniature of the whole story. The Open Group’s rationale states the divergence flatly — “The BSD echo checked the first argument for the string -n which caused it to suppress the newline,” while “The System V echo treated all arguments as strings to be written, but allowed escape sequences within them” — and then declines to pick a winner, ruling the overlapping cases implementation-defined and recommending printf instead. A standards body’s polite way of saying that a forty-year-old organisational disagreement is now permanently your problem.
The ps row has the same shape. Linux’s ps(1) documents three coexisting option grammars — Unix options that “may be grouped and must be preceded by a dash,” BSD options that “may be grouped and must not be used with a dash,” and GNU long options with two dashes — and warns explicitly that “commands options such as ps -aux are not recommended as it is a confusion of two different standards,” because a strict reading asks for processes owned by a user named x. The famous ps aux versus ps -ef trip hazard is not a quirk. It is two 1980s vendor blocs, preserved intact in a manual page.
Where Today’s Systems Actually Descend From
| System | Code lineage | Convention set | Certified UNIX? |
|---|---|---|---|
| macOS / Darwin | XNU — a hybrid kernel of Mach plus BSD (not a microkernel), inherited via NeXTSTEP, with much userland from FreeBSD | BSD | Yes. Apple ran the conformance suites and signed the trademark licence — ancestry alone would not qualify it |
| iOS, iPadOS, watchOS | Same XNU core | BSD | The registrations on The Open Group’s register are macOS’s |
| Linux | Clean-room. Descended from none of the three — the TUHS Unix Tree files it under “Clones,” systems implementing the interface without the original code | Both, plus GNU | No mainstream distribution; Inspur K-UX (expired 2019) and Huawei EulerOS have held UNIX 03 — economics, not capability |
| Android | Linux kernel, Bionic libc, no GNU userland | Its own | No |
| FreeBSD, NetBSD, OpenBSD | 4.4BSD-Lite | BSD | No |
| Solaris | SVR4 — Sun switched lineages: SunOS 4.x was BSD-derived, Solaris 2.0 (1992) was System V | System V | ⚠️ Absent from the register when it was read in August 2026; check before claiming it |
| illumos, OpenIndiana | OpenSolaris, therefore SVR4 | System V | No |
| AIX | System V-derived, with 4.3BSD-compatible extensions | Mixed | Yes (IBM) |
| HP-UX | System V-derived | System V | Yes (HPE, 11i v3) |
| z/OS | MVS lineage with a POSIX layer bolted on — not an AT&T descendant | POSIX | Yes (IBM). Certification tests conformance, not ancestry |
Two entries in that table carry most of the weight. Linux is Unix-like by interface and by nothing else. Torvalds wrote a kernel from scratch in 1991; it contains no Research Unix, System V, or BSD code, and saying “Linux descends from Unix” is not a simplification but an error — the canonical archive of Unix source itself classifies it as a clone. What Linux inherited was the specification-shaped part: syscalls, the file model, the process model, the command vocabulary.
And z/OS is the counterweight that proves the same point from the other side. It sits on the register as certified UNIX while descending from MVS. Ancestry and certification are orthogonal: one is about copied code, the other about a passed test suite and a signed trademark agreement. That separation was deliberate — when Novell transferred the UNIX trademark and the Single UNIX Specification to X/Open in 1993–94, The Open Group’s own history of UNIX describes the effect as separating the trademark from any code stream. It is why the same conformance question can be asked of a BSD descendant, a System V descendant, and a mainframe, and get a meaningful answer from all three.
How to Say It Without Being Wrong
| Don’t write | Write instead | Why |
|---|---|---|
| “System V 7”, “Unix V7 System V” | “Seventh Edition (V7), 1979” or “System V Release 4 (SVR4)” | Different organisations, different counters, twelve years and one antitrust settlement apart |
| “Unix version 4.2” | “4.2BSD” | BSD numbers are distribution numbers, not Unix version numbers |
| “BSD forked from System V” | “BSD began as add-ons to Research Unix and progressively replaced it” | BSD predates System V by five years and never branched from it |
| “Linux is a Unix descendant” | “Linux is Unix-like: it reimplements the interface, not the code” | No shared code with any of the three lines |
| “SVR4 was released in 1988” | “SVR4 was announced in October 1988 and reached products from early 1989” | Sources split on announcement versus availability — attribute it |
| “macOS is UNIX because it’s BSD” | “macOS is certified UNIX because Apple passed the conformance suites” | Lineage does not confer certification |
The compact version: an Edition counts manuals, a System V release counts products, and a BSD number counts tapes. If a sentence needs a version number in it, name the scheme in the same breath — “Seventh Edition (V7)”, “System V Release 4”, “4.4BSD-Lite” — and the ambiguity disappears.
The Lesson Under the Trivia
A version scheme is an artifact of an org chart, and all three of these are legible as such. Research numbered its documentation because a research group’s deliverable is a description. AT&T numbered products because a business unit’s deliverable is something a customer can buy and be supported on. Berkeley numbered tapes because its deliverable was a delta against a system somebody else owned — right up until the delta was larger than the original and the lawyers had to work out who owned what. Read anyone’s release numbering and you can usually reconstruct what they think they are shipping and to whom, which is a useful thing to be able to do to a vendor, a platform team, or your own repository.
The postscript is the one that matters most. All three schemes stopped. The Tenth Edition was the last; SVR4 was the end of the commercial line’s momentum; 4.4BSD-Lite Release 2 in June 1995 closed CSRG. What survived was the thing with no version number in this sense at all: the interface, written down, testable, and owned separately from any code stream. The artifacts were versioned. The specification is what got inherited — by Linux, which copied none of the code, and by a mainframe operating system that was never Unix in the first place.
❓ Frequently Asked Questions
Is “V7” the same thing as “System V”?
No, and this is the single most common mix-up in the subject. V7 is the Seventh Edition of Research Unix, dated January 1979, and produced by the Computing Science Research Center at Bell Labs; the number refers to an edition of the UNIX Programmer’s Manual. System V is AT&T’s commercial product line, whose first release shipped in January 1983 from the support-and-productisation side of the company. The “V” in System V is a Roman numeral five, not a version prefix — so “System V Release 4” carries two numbers, one naming the product family and one naming the release within it.
Which came first, System V or BSD?
BSD, by five years. 1BSD shipped on 9 March 1978 as an add-on tape for Sixth Edition Unix — a Pascal compiler and the ex editor, not an operating system. System V Release 1 shipped in January 1983. Both descend from Research Unix rather than from each other, which is why “BSD is a fork of System V” is wrong: they are siblings with a common ancestor in the Seventh Edition and its VAX port, 32V.
Does Linux descend from System V or BSD?
Neither. Linux is a clean-room kernel Torvalds began in 1991, containing no code from Research Unix, System V, or BSD. It inherited the interface — syscalls, the file and process models, the command vocabulary — which is a specification, not a codebase. The Unix Heritage Society’s own source archive files Linux under “Clones”: systems implementing Unix interfaces without the original code. Calling Linux “Unix-like” is precise; calling it a Unix descendant is not.
Why do both ps aux and ps -ef work on Linux?
Because Linux declined to inherit either convention exclusively. Its ps(1) implementation accepts three option grammars simultaneously: BSD options without a dash, Unix (System V) options with one dash, and GNU long options with two. That is a deliberate compatibility decision, not a standard — on FreeBSD, -e still means “display the environment,” so ps -ef does not mean there what it means on a System V box. The manual page even warns against ps -aux specifically, as a confusion of the two standards.
What exactly was SVR4, and why do sources give different years for it?
SVR4 was the deliberate merge of the two lineages: System V unified with 4.3BSD, SunOS, and Xenix, built jointly by AT&T and Sun. TCP/IP, sockets, UFS, and the C shell came from BSD; NFS, the virtual filesystem interface, the memory-mapped VM system, and the shared library model came from SunOS; x86 drivers and binary compatibility came from Xenix. It was announced on 18 October 1988 and reached customers inside vendor products from early 1989 — hence both years in circulation. Attribute the distinction rather than picking one silently.
Is macOS System V or BSD — and is it really UNIX?
macOS is BSD by convention and lineage: XNU is a hybrid kernel combining Mach with BSD, inherited through NeXTSTEP, and much of the userland tracks FreeBSD. It is also genuinely certified UNIX, unbroken since Mac OS X 10.5 Leopard in 2007 — but not because of that ancestry. Certification means Apple ran The Open Group’s conformance suites against a Product Standard and signed the trademark licence. IBM’s z/OS holds the same certification on an MVS lineage with no AT&T descent at all, which is the clearest available proof that “certified UNIX” describes a passed test, not a family tree.
📚 Further Reading
The primary sources here are free and worth reading directly. The Unix Heritage Society’s Unix Tree lets you browse the actual source of every version named in this article — Caldera released Versions 1–7 and 32V under a BSD-style licence in January 2002 — and Ritchie’s own Evolution of the Unix Time-sharing System is the best account of how the Editions actually progressed. For the standards half of the story, The Open Group’s History of UNIX and the freely readable POSIX.1-2024 base specifications are the documents that outlived all three version schemes.
- What Is a Container? The Idea That Powers Modern Software (and AI) — the direct descendant of
chrootand Plan 9 namespaces, and the modern place where “which Unix lineage am I actually running” stops being an academic question. - What Is Chromium? The Open-Source Engine Behind Chrome — exactly the same class of confusion in a modern code base: an upstream project and a shipped product treated as one thing because they share a name.
- Postgres as a Vector Database: What Benchmarks Say — Postgres came out of Berkeley alongside BSD and inherited its process-per-connection architecture from the same Unix design assumptions.
- Reading Code Is the New Superpower Skill for Engineers — the Sixth Edition kernel was small enough that a generation learned operating systems by reading it, which is precisely how the design propagated.
- How Tor Works: Onion Routing Explained for Engineers — a modern Unix daemon inheriting the privilege-separation model these lineages argued their way into.
📚 Recommended Resources
Disclosure: some links below are Amazon affiliate links. If you buy through them, I earn a small commission — at no extra cost to you — which helps keep this site running.
Start with the free primary sources. The Unix Heritage Society hosts the actual source of Versions 1–7 and 32V, released under a BSD-style licence in January 2002, so every claim in this article about what a given Edition contained can be checked against the code rather than a retelling. The POSIX.1-2024 base specifications — the document that outlived all three version schemes — are free to read in full. The books below are worth buying only once those have shown you what you want more of.
UNIX: A History and a Memoir
Brian Kernighan, 2019 · Book
The best single rung for the material in this article, and a primary source in its own right: Kernighan was in the Computing Science Research Center for the whole Editions arc and writes about it from inside the building. Read it specifically for the organisational texture — why the research group numbered manuals, what it did and did not consider its job, and how the commercial line came to be run by other people entirely.
A Quarter Century of UNIX
Peter H. Salus, 1994 · Book
The chronicle that pins down the sequencing this article compresses into tables — the Editions, the Berkeley distributions, and the AT&T commercial releases as three parallel threads, told with participant interviews. Out of print at the publisher, so this is a used-market purchase; Salus also authorised the Unix Heritage Society to distribute a scan, which is the free route if you would rather read it than own it.
The Design of the UNIX Operating System
Maurice J. Bach, 1986 · Book
The System V side of the split, from the inside. Grade its datedness honestly: it documents System V Release 2 specifically, which is exactly why it is useful here — it is a snapshot of what the commercial lineage’s kernel looked like before SVR4 merged BSD and SunOS work into it. Read as history and comparison, not as a current internals reference.
The Design and Implementation of the FreeBSD Operating System (2nd ed.)
Marshall Kirk McKusick, George V. Neville-Neil & Robert N. M. Watson, 2014 · Book
The BSD side of the same split, written by people who were in the CSRG for the events described here — McKusick’s participant account of the USL litigation and the 4.4BSD-Lite rebase is the standard source for that period. This is the current edition; there is no third.
Between them, Bach and McKusick let you read the System V and BSD lineages against each other at the level of actual kernel structure, which is where the convention differences in this article stop being command-line trivia and start being architecture.