4.3BSD

As with the 4.1BSD release, criticism was quick in coming. Most of the complaints were that the system ran too slowly. The problem, not surprisingly, was that the new facilities had not been tuned and that many of the kernel data structures were not well-suited to their new uses. Karels' and my first year on the project was spent tuning and polishing the system.

After two years of work spent tuning the system and refining the networking code, we made an announcement at the June 1985 Usenix conference that we anticipated releasing 4.3BSD later that summer. However, our release plans were brought to an abrupt halt by the folks at BBN. They correctly pointed out that we had never updated 4.2BSD with the final version of their networking code. Rather, we were still using the much-hacked initial prototype that they had given us many years earlier. They complained to DARPA that Berkeley was to implement the interface while BBN was supposed to implement the protocol, so Berkeley should replace the TCP/IP code in 4.3BSD with the BBN implementation.

Mike Karels got the BBN code and did an evaluation of the work that had been done since the prototype that was handed off to Berkeley. He decided that the best plan was to incorporate the good ideas from the BBN code into the Berkeley code base, but not to replace the Berkeley code base. The reason to retain the Berkeley code base was that it had gotten considerable testing and improvements from the widespread distribution in 4.2BSD. However, as a compromise, he offered to include both implementations on the 4.3BSD distribution and let users select which one to use in their kernel.

After reviewing Mike Karels' decision, DARPA decided that releasing two code bases would lead to unnecessary interoperability problems, and that just one implementation should be released. To decide which code base to use, they gave both to Mike Muuse of the Ballistics Research Laboratory, who was viewed by both Berkeley and BBN as an independent third party. After a month of evaluation, the report came back that the Berkeley code was more efficient but that the BBN code handled congestion better. The tie breaker was that the Berkeley code flawlessly ran all the tests while the BBN code panicked under some stress conditions. The final decision by DARPA was that 4.3BSD would stick with the Berkeley code base.

The polished 4.3BSD system was finally released in June 1986. As expected, it quelled many of the performance complaints, much as the 4.1BSD release quelled many of the complaints about 4BSD. Although most of the vendors had started the switch back to System V, large parts of 4.3BSD were carried over into their systems, particularly the networking subsystem.

In October 1986, Keith Bostic joined the CSRG. One condition of his employment was that he be allowed to finish up a project that he had been working on at his previous job, which was to port 4.3BSD to the PDP-11. While both Karels and I believed that it would be impossible to get a system that compiled to 250 Kbytes on the VAX to fit in the 64-Kbyte address space of the PDP-11, we agreed that Bostic could finish up his attempts to do so. Much to our amazement, the port was done successfully, using an intricate set of overlays and auxiliary processor states found on the PDP-11. The result was the 2.11BSD release, done by Casey Leedom and Bostic, which is still in use on some of the last remaining PDP-11's still in production in 1998.

Meanwhile, it was becoming increasingly obvious that the VAX architecture was reaching the end of its life and that it was time to begin considering other machines for running BSD. A promising new architecture of the time was made by Computer Consoles, Incorporated, and was called the Power 6/32. Unfortunately, the architecture died when the company decided to change its strategic direction. However, they did provide the CSRG with several machines that enabled us to finish the work, started by Bill Joy, of splitting the BSD kernel into machine-dependent and machine-independent parts. The result of this work was released as 4.3BSD-Tahoe in June 1988. The name Tahoe came from the development name used by Computer Consoles, Incorporated, for the machine that they eventually released as the Power 6/32. Although the useful lifetime of the Power 6/32 machine support was short, the work done to split the kernel into machine-independent and machine-dependent parts proved to be extremely valuable as BSD was ported to numerous other architectures.