Monday morning, Dr. S. Ramani, director of the National Center for Software Training, sent a huge old empty Tata bus to collect me. We rattled our way through the swarming vehicles adorned with "Horn OK Please" bumper stickers, past the shack which specialized in rewinding old fans, to the NCST headquarters.

My day with Ramani was a whirlwind of activity. Fascinated by how to use networks to help his country, Ramani rapidly jumped from ways to help the Indian Antarctic expedition to how to control infectious diseases to railway reservation systems to ways to disseminate Indian news to graduate students overseas.

He would leap to the white board and draw a diagram, then call out for his secretary to get certain staff members, then he would grab first one, then two telephones and start placing calls to officials and corporate executives for more information.

"Ramani, here," he said, calling the managing director of the Press Trust of India, a wire service with over 1,700 employees. "I have some ideas to discuss. Can you have dinner tonight at 9?"

The managing director had just gotten back from a trip to Delhi to meet with the Prime Minister, but dinner was arranged.

Meanwhile, we discussed ERNET, the Indian research network. The backbone was built around hubs in four cities–Bombay, Madras, Bangalore, and Delhi–which were connected to each other by 9,600 bps leased lines running TCP/IP. Other sites all over the country used UUCP into these hubs.

Over 70 sites in ERNET shared a single 9,600 bps satellite link to UUnet for international access. Needless to say, I felt more than a little bit guilty using Telnet to read my mail back in Colorado.

Low bandwidth, both domestic and international, was partly due to India's economic situation and partly to some technical detours towards satellites instead of terrestrial lines. Unlike some other countries I had visited, the underdeveloped Internet infrastructure was certainly not due to the lack of qualified engineers or a demand among users. India has one of the largest cadres of scientists in the world and the computer scientists I met at NCST were highly capable and fully aware of current developments in the Internet.

The economic situation was certainly a prime stumbling block in putting together a real research network (let alone a commercial service offering). The rupee had only just started to become convertible when I visited India and foreign exchange was at a premium. The U.S. half-circuit for the 9,600 bps line had been funded by the United Nations Development Program (UNDP).

Even within India, getting phone lines was no mean feat. Lines cost roughly the same as in the U.S., but average salaries in the U.S. are an order of magnitude higher than in India. The telephone infrastructure in India was notoriously underdeveloped, with electromechanical exchanges (i.e., something has to move to complete a circuit) still in operation in many locations. The wait for a residential line could easily take several years. Businesses pay a substantial surcharge to cut the wait to weeks or months.

In rural areas, lines are so unstable that running WCP over them could lead to endless retries. Even when the lines are operational, overhead of 500 percent is not uncommon. In fact, in some areas the lines were so bad that ERNET used a novel "floppy-based e-mail system." It turned out to be significantly more cost effective when the line got bad enough to simply spool mail to floppies and hire a courier to bring the diskettes down to the nearest hub. One system manager in Kanpur was once observed going through this biweekly ritual, neatly labeling floppies and stacking them into a pile.

"These guys have no other business," he grumbled, "all they do is keep sending mail through the day and night." Ramani showed me the network management utility which showed 30 to 70 messages exchanged per courier run with one of the FloppyNet sites in Kanpur.

Even in Bombay, local lines are not immune to problems. During the summer monsoon, 80 to 100 inches of rain can fall in 80 days of the monsoon season, knocking out service for 7 to 10 days per year. For this reason, the main international gateway had been located 20 km south of NCST in downtown Bombay. This machine, Sangam, was located in the Air India building and was just a kilometer from the telephone company's point of presence, making it fairly immune to disruptions during the monsoon.

A poor telephone infrastructure led Ramani and others in the Indian academic community to look to satellites, particularly the Indian manufactured INSAT series. The satellites gave high bandwidth, used Indian technology, and needed only a dish on the roof instead of depending on local infrastructure.

The original ERNET project plan from 1984 envisioned linking the 8 large academic and research institutions together with 4.5 meter dishes. For 1.5 million dollars, this infrastructure would provide multiple data paths, based on, of course, OSI protocols. By 1988, the project had evolved to specify 64 kbps data paths for the transmission of voice. data, and fax.

The voice requirement would satisfy needs for voice broadcast services for seminars, and would allow network management to have their own voice channels. The data channels would be established using the X.25 protocols and very coarse time slots of 1 second or even more would be provided. The data channels were thus suitable only for file transfer and electronic mail.

Although the project plans developed on schedule, the satellite kept getting delayed. By 1988, though, some Indian students returning from the states started agitating for Telebit modems, capable of running at fairly high speeds over voice-grade lines.

Anil Garg and some other NCST staffers began playing with UUCP dial-up transfers. The first links were in Bombay between NCST and a sister institution, the Indian Institute of Technology (IIT). At the time, the only phone available at the IIT computer science department was located in the office of the chairman.

To transfer mail, Anil would call up the chairman of the department, apologize for the interruption, and ask him to place the phone on the modem. If the phone was needed for voice sessions, then e-mail would have to wait. Eventually, a set time slot on the telephone was allocated and, assuming calls got through, e-mail started to flow regularly.

Over time, the terrestrial network started to grow. By 1991, with a little prodding from Vinton Cerf, an advisor to ERNET, the network was based on leased lines and Cisco routers. In early 1992, Ramani was in the midst of careful negotiations with the Department of Telecommunications to use a 64 kbps chunk of a new 140 Mbps fiber line to Delhi and maybe even to get a piece of the digital microwave links to Bangalore and Madras.

Meanwhile, the satellite project was finally getting off the ground, but its purpose had to be reevaluated. The terrestrial ERNET was obviously much more suitable for interactive applications and with 64 kbps terrestrial lines coming in the foreseeable future, Ramani was searching for a use for the satellites. Ramani was hoping that the satellite WAN could supplement the terrestrial network, perhaps being used by applications with a broadcast focus, such as the distribution of news.

Over a lunch of aloo bhindi and carrot halwa in the NCST canteen–equipped with a PC and custom software, of course–Ramani explained his attempts to automate the infectious disease units of the local hospitals.

Bombay hospitals get several hundred admissions per day for diseases ranging from dysentery to measles, polio, and hepatitis. An estimated 20,000 deaths per year are caused in Bombay by diseases that are preventable by vaccination, boiling drinking water, or other simple means.

When an admission occurs at a hospital, the admissions officer fills out a slip and places it in a pigeonhole for one of Bombay's 23 wards. The slips are picked up and, within 24 hours or so, make their way to public health officers in the wards who take appropriate actions. In the case of measles, for example, the health officer might go to a child's building and look for unvaccinated playmates.

Ramani had been conducting briefings and otherwise pushing people to computerize the process. He was hoping that a small number of computers could easily cut out the 24 hour delay. More importantly, aggregate information could be quickly examined for trends. A cluster of hepatitis cases, when displayed on a map would quickly indicate contaminated drinking water in a neighborhood. The information could be used to make special efforts in those neighborhoods to try and convince people to boil their water, at least for a while.

That afternoon, I FTPed my mail down to a local account to avoid tying up the international link, then went downstairs to give a lecture on the politics of standards. The audience of several dozen people were all active in implementing standards or using them as part of their work. This was certainly a sympathetic audience and everybody laughed when I explained the theory the standards potatoes advanced that anybody "serious" about implementing standards could certainly afford to participate in the process. Somebody came up to me after the lecture and explained how tough it was to get foreign exchange allocated for buying documents, let alone taking trips to conferences and standards meetings.

After the lecture, I went back to my hotel and pressed (having finally reprogrammed myself) the elevator's up button to go down to the coffee shop. The coffee shop was playing a rousing polka over the intercom, but I didn't see any kielbasas on the menu so I ordered some samosas and sweet lime water instead.

That evening, Dr. Ramani picked me up and we went to a local restaurant to meet Gourang Kundapur, managing director of the Press Trust of India (PTI). Over a dinner of brains masala and tandoori cauliflower, we gossiped about Indian politics, a topic even more complex than the politics of standards.

Tuesday morning, my driver took me from Juhu to Bombay to see the railway reservation system at Victoria Terminus, one of the main train stations and a Bombay landmark. Smack in the middle of rush hour, we dodged goats tied to shacks on the side of the road, pushcarts loaded with potatoes, and cows that ambled along content in the knowledge that they alone were safe on the roads.

My driver took what might charitably called a fairly aggressive approach to driving. He clearly felt that the green light was an indication that he should have already cleared the intersection. Even stuck behind several dozen cars, he would keep his hand on the horn. The side mirrors were carefully folded flat against the car, giving at least 3 inches of extra maneuvering room.

Arriving at the Central Reservations building, I entered a very large room filled with an incredibly dense mob waiting for tickets. Fighting my way up to the information counter, I was handed a form before I could even open my mouth, then was promptly swept along with the crowd.

The Indian Railways system is one of the largest in the world, moving 10 million passengers a day in over 6,000 trains. Tickets for these trains come in 7 categories (e.g., express or local), 32 kinds of quotas (e.g., foreigner or defense official), 100 types of discounts (e.g., veteran or handicapped), and 7 classes of reservations (e.g., first class or first class air conditioning).

The old, non-computerized system had, although I had a hard time picturing it, been known for incredibly long lines, which presumably meant that the lobbies must had somehow fit in even more people than I saw. To make a booking, you needed to be in the right line for the particular class of tickets for the particular train you wanted to take.

Demand has always far outstripped supply for Indian trains, so it was not unusual to have 1,000 people clustered around one clerk, while the next counter was empty. An 8-hour wait in line only to find there were no seats left was not unheard of.

Assuming you were able to fight your way up to the top of the line and get the clerk's attention, the process still had no guarantees. Complicated rate calculations, manual ledger books, and a host of paperwork led to lots of errors. When you got to the train, you might easily discover that there were duplicate bookings or that a supposedly sold-out train had dozens of empty seats.

To find out about the new and improved system, I met with officials from CMC, Ltd. the government-backed system integrators that had computerized the reservations process. CMC had set up five separate reservations systems for each of the five main regions of India, centered in Bombay, Delhi, Calcutta, Hyderabad, and Madras.

Four of the five systems were based on VAXen. For some reason, Hyderabad had some small CDC systems. Bombay was typical with two VAX 8650s and a 6310 clustered together. All the code was written in FORTRAN and even the database management functions had been locally developed.

The Bombay system handled 7 different reservation centers. Victoria Terminus was the largest, with 80 terminals, and the nearby Churchgate station had 20 terminals. Muxes and modems linked remote terminals to the cluster. A satellite office in Ahmedabad, 500 km away, linked 45 terminals over a series of eight 4,800 bps lines. All told, around 200 terminals handled 50,000 to 60,000 transactions on a typical day, with peaks up to 90,000 in the busy season.

For customers, the system meant that a reservation for any train in any class could be made in any line. Just as importantly, far fewer errors were made, since fares were automatically calculated and it was much harder to double-book seats.

Oddly enough, there was a downside. On the old system, it was possible to gauge your approximate chances of getting a ticket by the size of the queue and your ability to step over people and get to the front. On the new system, it was much harder to estimate your chances, since you competed with people all over the city. Monitors were posted at the stations indicating how many seats were available on the more popular runs, but the Chief Commercial Superintendent told me he was certainly receiving complaints from people unable to get tickets

The huge excess demand for tickets made scalping a profitable business. Ticket sales were limited to 4 to 6 per person to prevent scalping and each ticket had imprinted on it the name, age, and sex of the passenger. An ID was not required to ride the train, but conductors attempted to make sure that the holder of the ticket had at least some resemblance to the data on the ticket. The main result was that scalpers had adopted more sophisticated inventory management techniques.

While the seven main reservation centers in the Bombay region were online, smaller stations still used electromechanical teleprinters to communicate with Bombay. The messages would be received on slips of paper, a clerk would key in the data to the computer and send a message back to the station which would issue the ticket. Recently, data from a few teleprinter sites had been fed straight into the VAX. Hooking teleprinters to the VAX allowed properly formatted messages to be automatically answered. Not quite an interactive terminal, but much quicker than the old system.

This same message switch was being used to connect the independent systems for each region. When I visited, each center had a few terminals for each of the other regions, allowing a Bombay passenger to book passage on a train originating in Calcutta. Hooking those lines into a local VAX instead of a terminal would provide a way for one system to be a virtual teleprinter to a remote region.

After my briefing, my hosts insisted on showing me their machine room, which we entered via the washroom. Afterwards, I rode back out to Juhu, passing a line three blocks long waiting to enter the temple of Ganesh, the god with the elephant head.

Wednesday, Ramani sent one of the NCST jeeps to pick me up. The jeep was prominently adorned with "Govt. of India" on the front and back. NCST is not technically a part of the government, but Ramani explained that the labels made parking significantly easier.

I spent the day at the edge of the Ramani cyclone, periodically leaving to check on my FTP jobs, which were downloading source code for utilities that I thought might be useful to NCST. By the end of the day, netfind, perl, traceroute, and WAIS had all made it over the link.

The only thing I really wanted to bring down was the code for Cleveland Free-Net. A Free-Net is typically linked to Cleveland over the Internet and has a local bank of modems for public access. Ramani and I couldn't see how NCST could meet this requirement, but wanted to see if there was some way that the Free-Net concept could be applied in India.

Unfortunately, Free-Net wouldn't let us do that. Public access or nothing. People in India rarely had PCs and modems at home, and adding substantial traffic to an already saturated 9,600 bps link just didn't make any sense.

WAIS had no such arbitrary administrative restrictions, however. While ideally a WAIS client accessed servers all over the world, it was certainly technically feasible to run an isolated WAIS world on an Ethernet, providing a local information environment. Running WAIS with the NCST LAN was useful as a way for building up local expertise and seemed to fit right in with another current project.

NCST was taking the wire service from the Press Trust of India and feeding it into a VAX. There, the news feed was automatically broken up into individual items and fed into a local USENET news group. Every Friday, one day's worth of news was posted globally, providing information on Indian politics, sports, and culture to graduate students and professionals overseas.

Once the news hit the VAX, applications like WAIS might be an interesting method for searching and reading the news. NCST programmers were also developing their own sophisticated applications for sifting through information, such as a stock and news monitor that was built on SCO's Open Desktop.

All these fancy user interfaces made me curious how the news got produced. The next day, I went back into Bombay to visit the headquarters of the PII. We drove past a police officer on a chauffeur-driven scooter and ended up at Flora Fountain in the center of India's banking district.

The Uco Bank building, where PTI was located, was a few blocks away from where my driver left me off. He waved his arms vaguely in the proper direction, and I headed off down the street. Every block or so, I would stop a policeman or a rickshaw-wallah and say "Uco Bank Building" a few times.

I happened to be clutching a few random notes in my hand and every time I stopped to ask for directions, the person would grab the paper out of my hand and scrutinize it for several minutes. The notes had nothing to do with Uco Bank, but after a while the person would look up and wave me on down the street.

This method apparently worked because I soon saw a sign for the Uco Bank Building. I found it interesting that no matter which paper I had in my hand, it would be grabbed and examined. With a chuckle I thought of Cliff Lynch, one of the more active members of the library automation community. Cliff is famous for always clutching a stack several inches thick of business cards and scraps of paper. It would have taken him all day to get directions.

I went up the rickety stairs to the PTI offices and met Gourang Kundapur. Over cups of sweet coffee, he told me the history of PRI. Formed as a cooperative for small and medium newspapers in the early part of the century, the service was taken over by Reuters before the war. When the British left in August 1947, Reuters also pulled out and PTI was formed as a non-profit cooperative.

The news is gathered by over 400 journalists spread in 135 offices throughout the country. Before automation began, stories were typed on a teleprinter and punched to paper tape. The paper tape was fed in and the data went over a 50-baud line to one of four main centers.

In 1980, when Ramani began acting as a consultant to PTI, the entire network of 100,000 km of 50 baud lines was based on manual switching. The data came into a regional center and a tape was punched. A subeditor edited the story, and a new tape was be produced.

The tape then went to the transmission room, where an operator flipped switches to indicate which lines should be active, and the story was sent out. If the story had wide enough distribution, regional centers got the data, punched a tape, and sent it out to their local clients.

The obvious places to computerize were the four regional offices. PDP 11 systems were installed with custom code to terminate the teleprinter lines, acting as a switch and also providing an online editing environment for editors.

Connecting teleprinter lines turned out to be no mean feat. Each line transmitted at a slightly different speed. Custom boxes were developed that adjusted the speed to an exact 50 baud, manually at first and later automatically. The code from the PDP system was later ported to Xenix, with the Xenix system acting as an editing environment and a backup to the message switching function of the PDP.

While the regional offices were fairly well automated, Mr. Kundapur explained that it was not quite as easy to get rid of the teleprinters or even the 50 baud lines. The 50 baud lines were available to the press at 1/6 of the cost of normal lines, and, while it was possible to drive the lines at higher speeds locally, this would not work in remote areas.

People were another consideration. Electromechanical printers were noisy and dirty, but they had been used so long that people knew how they worked. Old-time journalists were not quick to adapt to electronic teleprinters (i.e., a dot matrix printer and a keyboard), although the mechanics tended to love the new systems as they required air conditioning.

For larger sites, teleprinters had been replaced with PCs or Atex systems. A custom device was built by a PTI subsidiary which adjusted the incoming voltage, changed the speed to 300 bps, and converted 5-bit BAUDOT code to ASCII, allowing most computers to easily accept a teleprinter feed.

PTI didn't have the most hi-tech system in the world, but it illustrated how to work in a technically challenging environment and still get work done. The system was continuing to move forward, with the central code being ported to an 80486 running UNIX and TCP/IP to link regional centers being investigated.

I got back to NCST in time for a quick lunch of baingan ka bhartha, a delicious eggplant dish similar to the Middle Eastem baba ganouj, accompanied by flat chapati bread. That afternoon and the next day I spent giving lectures on topics as diverse as WAIS and high speed networking and meeting with NCST students and staff. Saturday, Ramani and I got together to talk about standards.

Ramani was chair of a key committee in the Bureau of Indian Standards (BIS). One of the challenges in India was training large numbers of engineers, both for developing local systems and to stimulate the growing software export industry.

Training engineers meant giving them the information they needed and that meant giving them access to standards. BIS had been grappling for several years on how to distribute information, and online standards was high on their list. Ramani's committee had already decided that dissemination of information was a key priority.

Would Ramani be interested in posting standards for the Indian community? "In a minute," Ramani said without any hesitation at all.

This was the second place that felt that training people was much more important than playing international politics, and it was obvious that many other countries would post standards if the data were available.

It was obvious to me that I should get off the road and get busy preparing data, a task I couldn't do very well from an airplane. Before I could go home, though, I had one more stop to make.

I reported to Bombay's international airport for my 3:30 A.M. flight to London, Chicago, and Madison, Wisconsin. After seven stamps on my boarding pass, three in my passport, a special line and three stamps to clear my computer, a half-dozen luggage checks, and several lines with no apparent purpose, I was on my way.