How A Rocket Took Off: Juniper Founder Pradeep Sindhu (Part 3)

SM: When did you leave PARC, and why did you leave?

PS: In 1995, I realized I had been at PARC for 12 years and that I had tried every possible avenue I could find to take my ideas to the market. The difficulties that I faced were the same ones that ultimately led to the founding of other companies such as Adobe and Bay Networks. Xerox as a company understood paper and toner very well. It did not understand software and software systems. That was my area of interest, so I realized that it had come time to look for something else.

In the summer of 1995, I took a couple of months off with the intention of figuring out what to do next. I took a bunch of Red Herring magazines to my wife’s parents’ home in France to plan my next move. I looked at three different fields. One was in computing, and I did not see an area where I could make a difference. I looked in graphics in the image rending area, but the field was very crowded. Then I looked at networking. There had been plenty of networking culture at PARC.

In the networking universe I looked at the problem two different ways. One was in the LAN. I figured that this space was too crowded and that the problem was not complicated enough such that it really had room for another solution. In the WAN, the situation was much more interesting for many reasons. First, there was not a clear winner in terms of technology. Prevailing wisdom was some combination of TDM/ATM being the right way to solve the problem whereas IP was not recognized for its potential to do so.

I decided to look much deeper at the problem. My conclusion was that IP was actually much more scalable than ATM. The definition of the router is such that once you have the networking element, it makes it much easier to build scalable networks. The definition of ATM switches is that it is much easier to build them whereas networks were harder to build. They were not scalable.

If you look at an ATM switch and a router as a network element, the way that networks built with ATM switches are defined is that the model is actually still circuits. The essential difference is that before you can communicate end-to-end, you have to set up the circuit. That is followed by a communication and teardown phase. This three-phase model is actually perfect for telephony. I can understand why telecoms came up with this model.

Even back then, it was clear to me that in time, the bulk of the information that was going to go in and out of networks was going to be driven by computers. To take a model that was optimized for voice and phone calls, where an average call was four to five minutes long, was not ideal. Conversations between computers tend to be very short and have a lot of bursts. To me it was clear that ATM would not work.

Another fundamental reason that ATM was at a disadvantage is that in the ATM world, in order for a conversation to succeed, every element in the path has to work for the duration of the call. The probability of failure of the individual switches is p. The probability of the thing working is p raised to some power that is a very small number. On the other hand, if you take an IP network there are multiple paths to get from one place to the other. As there are circuits to set up, you can find an alternate path.

SM: What did Cisco and 3Com believe at the time?

PS: I do not have a good way of knowing what those companies believed internally. What was evident externally was from the products they built. 3Com products were on the Ethernet everywhere, even as a networking technology and not as an encapsulation layer. The notion that Ethernet was the right technology at scale was absolutely misguided and was eventually what killed 3Com. Ethernet is a great layer 2 technology, but it is not appropriate for layer 3 at all. When people have tried to apply it at scale they have failed.

For the wide area network, a key attribute is the ability to connect anything to anything at scale. The only technology I saw that was able to do that was TCP/IP. Another factor in play was the cost of raw transmission bandwidth. It used to be done with copper and was moving to optical fiber. That was dropping the cost rapidly. The cost of transmission was no longer a barrier. This meant that bandwidth supply was exploding.

This segment is part 3 in the series : How A Rocket Took Off: Juniper Founder Pradeep Sindhu
1 2 3 4 5 6 7