By "time-sharing" I mean an operating system which is

1. multi-user - more than one person can use the system at the same time.
2. multi-programming - more than one job can appear to run on the system at the same time.
3. interactive - a user can interact with the system without having to go through an intermediate step like submitting a batch job.
4. general purpose - the system can be used to develop and run programs in more than one language.

I also include virtualised systems like CP/CMS in the list.

We start with pioneers such as CTSS and Dartmouth DTSS in the early 1960s, but time-sharing really took off with mainframes in the late 1960s, with IBM's System/360 and Digital Equipment Corporation's PDP-10 having several commercial and academic operating systems. Also very influential was Multics, built by a consortium for a GE machine. IBM had many other competitors at this time, and outside the US several systems were developed. Minicomputers such as the PDP-8 and -11 started to appear towards the end of the decade, along with the birth of Unix on the latter machine. As the 70s continued, more minicomputers arrived, along with later large scale systems like PRIMOS and VAX VMS. Coming into the 1980s, focus shifted to microcomputers for personal use, with some outliers like MP/M and the workstation orientatted Domain/OS.

I have grouped operating systems into families where there is a clear line linking them together. This may span many years and different computer hardware. Probably the biggest example is Unix which is listed once in the table for the original implementation but went on to appear in many different versions (BSD, commercial Unixes like Solaris, Linux) across many different platforms.

As an operating system evolves, it can take on many names: for example the OS listed as CP/CMS includes its derivatives VM/370, VM/SE, VM/SP, VM/XA, VM/ESA, z/VM etc.

In some cases this is a judgement call - for example WAITS was originally based on TOPS-10, but I have included it as it is evolved to be a quite different system,

In scope are operating systems first introduced before 1983 - this is a fairly arbitrary cut off date chosen to focus on the period time-sharing systems were dominant before the personal computer started its ascent.

The date column indicates the first year an operating system was introduced: for commercial systems, the year of its first release to customers; for academic systems, the first year it ran production workloads.

Some operating systems had a lifespan of many years, gaining features and adding architectures. Some started as batch operating systems and had time-sharing added later. I have used the year when time-sharing features were added in that case. Some are still being maintained today, albeit in quite different forms from the original version, such as MVS, MCP and VMS.

This column was by far the hardest to verify and I would appreciate any correctiobs.

One complication then is, when we look at an operating system which version should we select? Often the choice is made for us, so for CTSS we have a single source from around 1969; in other cases like TOPS-10 we have a choice of releases spanning 1972 to 1988.

I will generally choose to look at the most recent version of a system, but this does make a chronological comparison of different systems more difficult.

I concentrate on running operating systems under emulation here, which anyone can do if they have a personal computer. The gold standard is an open-source, cross platform emulator, but I do included single platform ones (eg emulators running on Windows only) and closed-source but freely available emulators where this is the only choice. I do not include commercial, paid-for emulators as the focus for this site is hobbyists and researchers.

Some systems have more than one emulator. I usually choose simh or hercules if available, if not the most fully features single architecture emulator. In some cases where there is sufficiently different functionality or approach to emulation I list more than one.

simh is the most frequently represented emulator on this list as it emulates more than 30 different architectures. There are several different forks, but in most cases a recent version of open-simh is a good choice.

hercules also appears several times due to the large number of IBM System/360 operating systems. Again there are several forks, but SDL hercules hyperion is my current preference.

I have not (yet) run all the emulators listed in the table, so it's quite possible some do not work fully - please let me know if so.

Alternatives to emulation running on your own personal computer:

• hardware for more recent systems such as PDP-8s or VAXes is available on sites like ebay, but is getting increasingly expensive.
• replicas running emulators on hardware like Raspberry Pis, such as the PiDP-11.
• some museums has put original hardware online where you can get an account, such as The Interim Computer Museum.
• there also collections of online-accessible emulators such as nostalgiccomputing.org.

In the "SW" column I link to the project page or archive site where software kits can be obtained. Be aware that some kits have specific licenses attached and others have no obvious current owner so their licensing status is unknown.

The "Doc" column links to the main source of documentation for the operating system in PDF format: this is usually Bitsavers.

This is not a comprehensive list of all time-sharing operating systems ever produced - to my knowledge, no such list exists but I estimate it would be at least double this size.

In order for a system to be available on emulation, several things need to happen (or to have already happened)

• the original software and documentation needs to have been preserved, ideally in digital form
• sufficient details of the hardware it ran on need to have been preserved
• someone has made the big effort to write an emulator for the system

Some examples of systems that are not available, at least today, via emulation

• early systems like AN/FSQ-32, JOHNNIAC or the PDP-1 systems developed by BBN and MIT, along with British systems like the Titan Supervisor
• Mainframe operating systems like GE GCOS, RCA TSOS, Univac Exec 8 and ICL VME.
• Academic systems like CAL-TSS, OS-3, CAP or EMAS.
• Many minicomputer systems.
• More recent systems that are only available commercially, eg Tandem or Stratus VOS.

And in many cases, even for those available, not all versions of each operating system has been preserved.

Please let me know if you think I am missing any systems, or if you think my methodology is not right. You can email me at rupert@timereshared.com and if it makes sense I will add it here and update the main text.

August 2025: Lars Brinkhoff suggested adding SITS.

Algol was originally designed by a committee of European and American computer scientists in the 1950s-60s. The first version to be widely implemented was Algol 60, which is what we will look at here.

Dartmouth had experience of implementing subsets of Algol on its smaller computers before DTSS, and this was the second language to be introduced on DTSS after BASIC in 1964.

An illustration of a GE-225 system from "GE 200 Series Operation Manual", GE, 1966. Source: bitsavers.

One important point when dealing with Algol is different levels or representations of the language: there is one for the spec, one for printed programs and one for the machine implementation. As an example, to raise a to the power b:

• Spec: a ↑ b
• Printed: a^b
• Machine - Dartmouth: a ^ b
• Machine - IBM: a ** b

We'll use the printed form in this article and the Dartmouth machine form for code examples.

Keywords are in bold lower case in the printed form (eg integer). Dartmouth uses upper case (eg INTEGER). (IBM used quotes, eg 'INTEGER', which must have been a pain to type).

Here's an example program that will print Hello World 5 times

10 BEGIN
20 INTEGER I;
30 FOR I := 1 STEP 1 UNTIL 5 DO
40     PRINT("HELLO, WORLD")
50 END PROGRAM

The line numbers are not part of Algol but are needed for the Dartmouth program editor.

Statements are grouped into blocks, separated by semicolons; the last statement in a block does not need a semicolon.

Variables are typed (real, integer, boolean); strings are allowed only as constants.

Algol has no standard I/O - reflecting how different each computer was at that time, it was up to the implementation to add this. Dartmouth Algol uses print much like BASIC for output.

Let's exercise Algol further by implementing the TPK algorithm.

100 BEGIN
110     COMMENT TPK ALGORITHM IN ALGOL 60;
120 
130     REAL PROCEDURE FN(X);
140     VALUE X;
150     REAL X;
160     BEGIN
170         FN := SQRT(ABS(X)) + 5*X^3
180     END PROCEDURE;
190 
200     COMMENT MAIN PROGRAM;
210     INTEGER N, J;
220     REAL ARRAY A[1:11];
230     N := 11;
240     COMMENT READ NUMBERS FROM DATA;
250     FOR J := 1 STEP 1 UNTIL N DO
260         READATA(SAMPLE, A[J]);
270     PRINT("RESULTS ARE");
280     FOR J := N STEP -1 UNTIL 1 DO
290     BEGIN
300         REAL RESULT;
310         RESULT := FN(A[J]);
320         IF RESULT > 400.0 THEN
330             PRINT("TOO LARGE")
340         ELSE
350             PRINT(RESULT);
360     END LOOP;
370     DATA SAMPLE := 10, -1, 1, 2, 3, 4, 4.3, 4.305, 4.303, 4.302, 4.301;
380 END PROGRAM

The indentation is for readability only and is not required by the language. Contemporary users would probably pack more statements per line so as to use less vertical space on the typewriter output. It's not shown here, but you can also spread out statements over several lines and even include spaces in variable names.

In the procedure definition, note that we declare the input parameter X as having real type and also value class, so it is passed by value; the other option is name which is similar to pass-by-reference.

The main program starts on line 200. Note that variables need to be declared before they are assigned to, and arrays can have arbitrary indices.

The program uses the readata procedure, which is a Dartmouth extension to read items from the matching data statement, similar to BASIC READ and DATA. print is also a Dartmouth extension.

At line 300, note that RESULT is declared in the for block so its scope lasts until the end of the block only; this was an innovation of Algol.

• See the quick tour on how to get the emulator running.
• Download the source file from Github.
• If you are not on Windows, ensure the file has CR/LF line endings. A simple way to do this:

awk 'sub("$", "\r")' < input_file > output_file

• Store the file as TPKA (no extension) in the same directory as DTSS.EXE.
• Start DTSS.
• Type SYSTEM ALGOL to switch from BASIC to Algol.
• Type OLD TPKA.
• Type LIST to confirm the source has been loaded.
• Type RUN.

The DTSS zip file contains An Algol Outline.pdf.

The Revised Report on the Algorithmic Language Algol can be considered as the standard definition of the language.

There are some contemporary books on the Internet Archive: Dijkstra's A Primer of Algol 60 Programming is terse but readable; McCracken's A Guide to ALGOL Programming is a more traditional text book.

I wrote an overview of Algol 60 on the IBM S/360 under MTS here, which goes into more detail on the language and how IBM implemented it.

I welcome any questions or comments, and also especially any corrections if I have got something wrong. Please email me at rupert@timereshared.com and I will add it here and update the main text.

There are only 15 keywords.

    LET       GOTO      RETURN 
    PRINT     IF        DEF    
    END       FOR       DIM    
    READ      NEXT      REM    
    DATA      GOSUB     STOP

All variables are floating point numbers, and variables names must be a single character or a single character followed by a single digit. You have to use the LET keyword to assign to variables.

PRINT can print numbers or strings enclosed in double quotes.

This early version could only write data back to the user running the program and was unable to take input from the the user. This was implemented two years later by adding the INPUT keyword and OS support. The emulator does support INPUT, but we'll be historically accurate by using the READ statement to take data from a DATA statement in the program.

GOTO allows jumps to any line number, GOSUB also jumps to the given line number, but when RETURN is reached it will transfer execution back to the line after the GOSUB.

Up to 26 functions, containing a single expression only like Fortran, can be defined with DEF.

The result of an IF statement can only be a line number which is jumped to if the expression is true.

Let's exercise BASIC by implementing the TPK algorithm.

100 REM TPK ALGORITHM IN BASIC
110 
120 DEF FNT(X) = SQR(ABS(X)) + 5*X^3
130 
140 REM MAIN PROGRAM
150 DIM A(11)
160 LET N=11
170 REM READ NUMBERS FROM DATA
180 FOR I = 1 TO N
190 READ A(I)
200 NEXT I
210 PRINT "RESULTS ARE"
220 FOR J = 1 TO N
230 LET K = N - J + 1
240 LET R = FNT(A(K))
250 IF R > 400 THEN 280
260 PRINT R
270 GOTO 290
280 PRINT "TOO LARGE"
290 NEXT J
300 DATA 10, -1, 1, 2, 3, 4, 4.3, 4.305, 4.303, 4.302, 4.301
310 END

In line 120, we use ^ to raise X to the power of 3.

Arrays are zero indexed, but we use 1-11 for the array defined on line 150.

The READ on line 190 takes values in order from the DATA on line 300. You could substitute INPUT A(I) here to take input from the user.

• See the quick tour on how to get the emulator running.
• Download the source file from Github.
• If you are not on Windows, ensure the file has CR/LF line endings. A simple way to do this:

awk 'sub("$", "\r")' < input_file > output_file

• Store the file as TPK (no extension) in the same directory as DTSS.EXE.
• Start DTSS.
• Type OLD TPK.
• Type LIST to confirm the source has been loaded.
• Type RUN.

The results will look like this:

OLD TPK
RUN

TPK     18:55      JUN 23, 2025

RESULTS ARE
 399.886
TOO LARGE
TOO LARGE
TOO LARGE
  399.609
  322
  136.732
  41.4142
  6
-4
TOO LARGE


TIME 0. SECS

The DTSS emulator zip file contains A BASIC Outline.pdf. A 1964 introduction to the system and BASIC can be found on bitsavers.

I welcome any questions or comments, and also especially any corrections if I have got something wrong. Please email me at rupert@timereshared.com and I will add it here and update the main text.

The user interface was much like we saw in the emulator quick tourm with the addition of a HELLO command for the user to identify themselves via a numeric ID; there were no passwords.

The user can either be editing their program - indicated by a typing a number at the start of a line - or submitting commands such as LIST or RUN. The commands are not tied to BASIC, so you could use them if programming in ALGOL, for example.

The DATANET-30 was intended by GE to be a smart communication controller, capable of handling several terminal lines, buffering data and sending messages. It had 18-bit words and was relatively fast. It was interrupt driven and had a stored program facility which Dartmouth used to create an operating system. This was divided into real-time and spare-time tasks.

The real-time part was driven by an interrupt running 110 times per second. It would scan the lines for activity and store them in buffers. For commands, it would set up a job entry and store it in the spare-time queue.

The spare-time part ran every second and would examine its queue. Programs being developed would be sent to the disk. Command such as RUN would cause the DN-30 to send a message to the GE-235 to do the actual work of compiling and running the program.

The disk drive stored up to 16 million words of data and was connected to both the DN-30 and the GE-235.

The GE-235 was more powerful than the DN-30 for integer and floating point arithmetic. It had up to 16,000 words of core memory, each word having 20 data and 1 check bit.

During time-sharing, its operation was controlled by the DN-30. It held the compilers and execution services in memory. When it receives a message from the DN-30, it read the user program from the disk and starts compiling and executing it - so user programs were compiled every time they were run rather than saving object files. At run-time user programs had access to 6,000 words of memory.

As far as I can tell from the documentation, only one user program could be in memory at any time: the DN-30 could tell the GE-235 to write the contents of memory to disk in order to process another user program. The 1964 documentation states that the system was good at running a number of short jobs this way, but there could be waits of 5-10s for jobs to be executed. However, longer running jobs were not really suitable for this system.

Peripherals such as tapes and card readers were not used during time-sharing, but were used for system development purposes, presumably in a batch execution mode.

From bitsavers:

• The 1964 DTSS System Description from Dartmouth.
• The DATANET-30 and GE-225 programmer's reference manuals from GE.
• A book from 1980 covers later versions of the operating system.

I welcome any questions or comments, and also especially any corrections if I have got something wrong. Please email me at rupert@timereshared.com and I will add it here and update the main text.