Viewing notes. The pictures are 'tear-off'. Line 'em up left to right for a more-or-less continuous timeline. There are four threads that follow through --
I picked 1968 as a place to start because that's the year DCA was formed to converge the (perceived) duplication of long haul communications infrastructure between Army and Navy.
It's also a convenient place to start because the first four ARPANet hosts were hooked up in 1969. And shortly after, DARPA forms, and supports, the Internet Engineering Task Force both as a community of interest to formulate the protocols and as a means of documenting them.
1969 is also a good year since that is the year that Brian Kernigan and Dennis Ritchie release the first publicly known version of Unix. This work is being done at AT&T Bell Labs (now known as Bellcore). Because AT&T is a phone company, and effectively a regulated monopoly, it is not allowed to sell products in the computer industry. As a way out of this dilemma, DARPA forms the Berkeley Software Design team to advance the development of Unix and also to make licensable products available to universities (non-commercial).
Prior to 1968, the telephone company was treated as a regulated monopoly -- top to bottom. Long distance, central office switching, local loop and user equipment (telephones) were all available only from the phone company. The Carterphone decision was one of the significant unravellings in the user equipment category. Carterphone was essentially in the telephone-patch business and tying the earphone of the telephone to the microphone of a business band radio in order to provide telephone services to offshore oil rigs. AT&T sued Carterphone, alleging an 'incompatable' connection to the telephone system. AT&T lost on appeal, leading to the situation today where you can buy a telephone instrument at the local drugstore, take it home and plug it into the jack on the wall. We went through the same drill in the early seventies with telephone modems.
The mid '70s is a period of technological development. Most of the interesting infrastructure development comes later.
DARPA is still steadily supporting IETF. Meanwhile, out in industry, the Unix aficianados are getting hooked -- they are using Unix in school and want it when they get out in industry. UUCP is Unix-to-Unix CoPy -- rather a rump bottom-up network development.
Meanwhile a young company named Intel produces what it perceived as the next generation calculator chip -- the 8080. Some imaginative guys in garages started building computers with the chips -- something qualitatively different than a calculator.
The turn of the decade sees some significant maturations in the technology which presage some major growth. In the ARPANet protocols, the Network Control Protocol is replaced by a pair of protocols, TCP and IP, which elegantly conform to our perceptions of the Transport and Network layers respectively.
NCP, because it tried to perform both functions, lacked scalability. With some perception, you could also see some significant software maintainability problems coming up too.
The TCP/IP bifurcation is a technological enabler in Internet growth because we lay the technical foundations for IP (and kin: ICMP, RIP et al) to reside in routers which are stateless traffic directors. TCP is the state machine and does not reside in routers at all, rather this is the functionality that lives in end systems. This decoupling allows each part of the infrastructure to grow independently.
Meanwhile, the DARPA-supported BSD project is chugging along. With the 4.1 release, the TCP/IP protocols are bundled in the operating system -- a major reason why Unix machines routinely network cleanly ... something that PC vendors took another decade to catch on to.
And the BSD project is spawning more than just software. Scott McNealy, Bill Joy and some others are graduates/disciples of the Berkeley work and go off to found Sun Microsystems which ships commercial workstations with licensed copies of ... BSD Unix. Most of Sun's competitors (Apollo, for example, later bought by HP) are trying to use proprietary operating systems. Sun, with Unix, quickly comes to dominate the workstation market. (Apollo's fortunes are saved by the HP purchase a few years later, which also sees a clear shift -- HP ships the Apollo workstations with HP-UX, their version of Unix).
This is also the point where the federal government breaks the phone company monopoly in long distance service. The Modified Final Judgement, by Judge Greene, breaks AT&T into a parent company (AT&T) which retains the long distance service and 7 Regional Bell Operating Companies (RBOCs). The research establishment remains intact (Bellcore) but put on an arms-length contractural basis with the rest of the industry. Note that the central office switching and the local loop copper plants are still effectively monopolies ... and rather beyond the reach of the FCC as rates and access are controlled by the state Public Utility Commissions.
In this timeframe we see the beginnings of some major scale-up. The technology works. It's mature enough to use in the real world. Now make it grow.
In 1984, ARPANet is declared a success by DDR&E. The competitor development, AUTODIN II, is cancelled. ARPANet is split into two parts. The university portion remains ARPANet, but is rapidly evolving from an experimental network to a utility that supports other research. See the next panel for what happens to this part. The operational portion is renamed Defense Data Network which is in turn split into four parts, segregated for security reasons:
DCA shoulders responsibilities that we today call those of an Internet service provider.
We also see a major piece of governmental schizophrenia developing. DARPA is still supporting IETF, which is dutifully producing ARPANet standards which are actually getting implemented. The process is extremely informal, by standards-body norms; this lack of orthodoxy has image problems in government.
Once DDN is declared as a 'real' network, its management shifts from DARPA to DCA. DCA is not impressed by the IETF informality and instead latches onto the much more formal networking standards process in ISO. The DCA Center for Standards pushes NIST to develop the Government Open Systems Interconnect Profile -- the ISO OSI standards packaged for federal government use. (85% of the NIST GOSIP budget comes from DoD).
Two interesting developments in industry deserve note. Bob Metcalfe and his team are developing Ethernet as a LAN standard. There are several also-rans in this period, but Ethernet becomes dominant in the marketplace and eventually gets anoited with an IEEE standard (802.3). Of this generation of LAN standards, only the 802.5 token ring survives besides Ethernet. To my knowledge, none of this LAN technology development has any major federal funding -- it's all commercial industry work. Bob Metcalfe was a graduate student involved in ARPANet before he went on to develop Ethernet.
The other development is Phil Karn writing an implementation of TCP/IP that works on DOS PCs. (Phil's amateur radio call is KA9Q). Phil is working at Bellcore at this point. The technological trick is that TCP must multi-task if it is to be effective ... and DOS doesn't multi-task. So Phil wrote the multi-tasking within the application -- a readily admitted kluge, but it worked. Phil licensed the code but made it freely available for non-commercial use; it quickly took over the amateur packet radio community. Phil's code also got licensed and found its way into several other places, like Telebit's Netblazer.
The amateurs performed a very valuble service in the way they supported Phil. Like the sanitation department that follows the horses in a parade, they followed Phil around and wrote up the documentation. Nice piece of teamwork.
DARPA is still supporting ARPANet, but is in the process of getting out of that business. ARPANet has ceased being an experiment and has become a utility within the university environment -- and support of utilities is not the business DARPA is in. NSF is in that kind of business though, an the arrangements for the shift in management are completed at the end of this period.
The late '80s sees a sizable critical mass appearing. The foundations in both technology and infrastructure have been layed over the past decade and now the 'net hits its stride.
Within DoD, 'net use is increasing significantly. The problem is with DCA (now being renamed DISA). DDN access is expensive, and slow to arrive. From the time you could document requirements (flag signatures on formal letters), you spent two years in the User Requirements Data Base before installation, and DDN's chargeback schedules were nowhere near keeping pace with the declining costs in what is turning into a commodity business. Besides, few people in the military services trust DISA to deliver anything. So a lot of enterprising folks out there start building networks. A part of the Navy starts buying its own T-1 sized pipes (24 times larger than a 56k line DDN wants to furnish you), routers, and hooking up its own network.
I'd suggest that this isn't all bad. The bypass networks are still using TCP/IP so they are fully interoperable with the DDN connectivity -- making them fully in consonance with the original DCA charter.
Over in NSF, the strategy is to make the Internet (renamed ARPANet) grow from k's of users to M's of users. NSF contracts with ANS (consortium of IBM, MCI and MERIT) to operate the backbone. NSF also encourages regional university consortia to set up Research & Education networks that interconnect to the backbone. This encouragement takes the form of a subsidy to offset the setup and operation costs -- a subsidy that is fully phased out 5 years later.
The Internet has been growing at a rate exceeding 100% per year ever since 1969, but in this period the numbers get somewhat larger, causing some interesting scaleup problems. (ask your stockbroker...). Several protocols reach their limits: Routing Information Protocol can only reach destinations that have fewer than 15 routers between them ... and the net now has diameters larger than that -- unthinkable in 1970. Similarly, Internet Protocol itself is approaching limits imposed by the 32 bit address space and very inefficient assignment processes.
Also during this period, the Internet becomes a consumer item. One had to be a geek to wrestle with the e-mail implementations, ftp, telnet, etc. Consider this analogous to automobiles before the Model T -- cars were rather custom-made, and automobile users had to know a lot about their contruction and repair because they had to do it themselves. World Wide Web and browsers like Mosaic (and Netscape) represent to me the Model T -- this is the beginning of Internet for everybody -- even flag officers can use this stuff!
Partly from pure growth of the Internet itself and partly prodded by the end of the NSF subsidies, the Internet transforms itself in this period from a research & education (non-profit, non-taxable, non-commercial) entity into a revenue-producing part of the US economy. Our tax dollars spent by ARPA and NSF have served to guide and seed an entire industry that now far outstrips the relatively modest government investments.
Where to next? In a couple years, I'll build the next panel.
But I'll make some predictions: