BeBox History
Be Newsletter Issue 3
December 20, 1995
BE ENGINEERING INSIGHTS:
"Our" Toy Story
By Steve Horowitz
The BeBox you see today is actually quite a bit different than the one that
existed four years ago. Both the hardware and software have come quite far
since the version we had running in our old offices, in an industrial park in
San Jose in 1991.
The original design of the BeBox was actually implemented as a dual-Hobbit
processor machine, with three digital signal processing (DSP) chips on the
motherboard. The bus architecture was proprietary, but memory, hard disks,
mice, and keyboards were all borrowed from what Jean-Louis likes to call the
"PC clone organ bank."
The software at that time consisted of just a multiprocessing, multitasking
kernel and a command-line shell for launching applications. There were no
graphics, no windowing system, and no integrated database -- although all of
these items were planned.
The first attempt at bringing some graphics capabilities to the machine came in
the form of an evaluation of the NeWS windowing system from Sun Microsystems.
My first job at Be was to learn PostScript to see what it was like to program
in the NeWS environment. While NeWS did offer us a client-server approach to
graphics and PostScript as a graphics model, it was rather large and the
programming model, as well as the port to the BeBox, seemed complex.
While this evaluation was taking place and while we continued to evaluate other
user-interface tool kits and graphics systems, an engineer named Benoit
Schillings was quietly developing his own graphical windowing system. Using his
intimate knowledge of the Be kernel, he optimized his graphics system to run
extremely fast on the BeBox. To this day, Benoit remains our primary resident
"speed freak."
The decision at that point became obvious. Although writing the entire graphics
system from scratch was never our intent, we could see that it was the only way
to achieve the kind of responsiveness and usability that we wanted for the
BeBox. I abandoned the evaluation process and set out to help Benoit write the
first version of the Be class library, which would interact with his speedy
graphics server. Using the initial classes we developed (including windows,
views, scroll bars, buttons, and menus), we set out to write a number of
applications, including the file system Browser, to demonstrate and test the
system. We were later joined by Eric Knight, who worked on the class libraries
as well as several key demo applications, such as MIDI, MessageCenter, and
FontDemo.
While all of this "high-level" work went on, the kernel, file system, I/O
system, and device drivers continued to evolve and improve under Erich
Ringewald, Bob Herold, and Cyril Meurillon, who had all been with the company
since its inception.
In early 1994, Peter Potrebic joined the Be team and was given the
responsibility of maintaining and enhancing the class libraries. Under Peter,
the application framework saw a number of significant enhancements, including
an improved application messaging system and the addition of threading classes.
During release 2 of the software he was absolutely zealous in tracking down a
number of extremely difficult bugs, which resulted in the system reaching a
level of stability that we had never seen before. Our technical writers, Don
Larkin and Doug Fulton also helped tremendously to make sure that the design of
the class libraries and APIs were coherent and consistent.
Shortly after Peter joined the team, we hired another engineer named Robert
Polic. Robert started out by helping Benoit with the graphics server and then
showed tremendous versatility as he went on to work on many different levels of
the system, from the SCSI and video drivers to many of our utility
applications, such as IconWorld, CDPlayer, Keyboard, and all of our preferences
applications.
By this time the team had grown large enough that certain development tools,
such as source control and a bug-tracking system, became essential. Ming Low,
who had been in charge of all of our builds, developed both of these tools and
many others to keep the development effort running smoothly as the team grew.
As the team grew, so did Ming's responsibilities, which included system
configuration, network debugging, testing, and helping with each release of the
software.
All of our software, of course, was still running on the dual-Hobbit, 3-DSP
machine. While we were questioning the wisdom of having specialized DSP chips
on the motherboard, we saw the entire future of the hardware cast in doubt when
AT&T canceled the Hobbit processor line. It was decision-making time at Be
again, and this time we saw the opportunity to make improvements in several
areas. We needed a new processor and the PowerPC line seemed appealing on a
number of fronts. First, it offered pretty good performance for the price.
Second, it had the backing of a number of industry leaders (Apple, IBM, and
Motorola) who assured its existence and design evolution. Last, but not
insignificant from our perspective, was that unlike the Hobbit, the PowerPC
chips had the ability to handle floating-point math very quickly. We saw this
as our opportunity to abandon the specialized DSP chips and gain a much simpler
system design. This new design retained the ability to process floating-point
numbers very quickly, and was also able to throw both PowerPC chips at general
system tasks if needed.
The entire PowerPC motherboard design, including the use of industry-standard
ISA and PCI buses, was accomplished by a single hardware engineer, Joe Palmer.
The port of our software to the PowerPC processor was basically the work of two
engineers, Bob and Cyril. Bob worked magic as he single-handedly brought up
each revision of the new hardware and debugged impossibly complex interactions
between the lowest levels of our software and the new hardware. Cyril,
meanwhile, rewrote a number of speed-critical, machine-dependant pieces of the
kernel, including the interface to the memory management unit. The rest of us
"high-level" guys basically just continued our work on the Hobbit boxes until
the day Joe showed up in our offices with new PowerPC machines. With a new
compiler and literally a simple recompile we had all of the applications,
including the Browser, running on the new PowerPC boxes.
There was another benefit from the switch to the PowerPC, which we came to
appreciate a great deal. Having designed all of our software for the Hobbit we
were forced to spend a lot of time looking at performance issues, which we
might have ignored had we started with the faster PowerPC. This resulted in a
system that ran very well on the Hobbit processors, but was noticeably snappier
on the PowerPCs.
As the PowerPC design took hold we added another key engineer, Brad Taylor, who
quickly became the most popular guy on engineering row when he implemented the
TCP/IP networking stack on the BeBox and ported the ftp tool for us to use for
file transfers. Before Brad ported ftp, we had agonizingly used the tar tool to
transfer files on floppies from our UNIX development machines to our BeBoxes.
With ftp, Brad turned my transfer time for the Browser from one minute to three
seconds. Since most of you reading this are developers, you know what a
tremendous difference this makes in the development process.
The team itself was run by our VP of engineering, Erich Ringewald. Erich's
contributions to the BeBox are considerable. Along with writing kernel code,
keyboard drivers, and porting UNIX tools he has kept the team on track
throughout its existence. He has refereed fierce battles in the engineering
team and has made critical decisions throughout the design process. Without
actually micromanaging any individual, Erich has managed to keep a team with
extremely diverse skills and personalities all focused on the same goal.
Some of the other later additions to the Be product team have been Melissa
Rogers, Rico Tudor, Roy West, and Guillaume Desmarets. Melissa is the
self-described "chief pitbull." She keeps the engineering team in line and is
responsible for the project schedule and about fifty other things at Be. Rico
was given the task of porting the bash shell to the Be environment and writing
a new ansi terminal application to host bash. Roy thought he was just going to
write user documentation, but has ended up becoming our webmaster, writing and
editing marketing materials, and proofing this newsletter. And he's even
managed to write some user documentation. Guillaume has joined us to help Joe
with future versions of Be hardware.
I hope this article has given you a feel for the evolution of the BeBox
hardware and software and for some of the major milestones along the way. Even
though we've worked on the BeBox for a number of years there's still a great
deal left to do. With your help and feedback we'll continue to improve the Be
hardware and operating system to give your software a solid base to run on. I
hope to write a similar article in the coming years that will continue the
story, with all of the incredible innovations you'll bring to the BeBox in the
form of enhancements, utilities, and ground-breaking applications.
|
