|
Be Newsletter
Issue 8, January 31, 1996
Table of Contents
BE ENGINEERING INSIGHTS: The
Be OS from a UNIX Perspective
By Dominic Giampaolo
If you come from a heavy UNIX background like I do, you may
be wondering just what the Be OS looks like under the hood
and how it compares to the legendary heavy-weight of big-boy
operating systems.
As a starting point, let me say right off the bat: The Be OS
is not UNIX nor is it intended to be a UNIX clone. That's
not to say that there is no UNIX compatibility, as we do
have a Posix library that's intended to make it easy to port
UNIX application code (and release 1.1d7 will offer a big
leap forward in that area). However, the fundamental
abstractions provided by UNIX and the Be OS are
different.
Let's begin with the model of a program. Both the Be OS and
UNIX offer separate address spaces for each "team"
("process" in UNIX-speak). Each program runs in its own
2-gigabyte memory space and can use as much or as little of
that space as it needs, since the Be OS (like UNIX) offers
virtual memory. Separate address spaces of course mean that
each program is isolated and cannot accidentally rampage
through memory.
Unlike most versions of UNIX, the Be OS offers multiple
threads of execution within each team (address space). To
create a thread, the Be OS provides spawn_thread().
To create a new team, there's the function
load_executable() (which also creates a default
thread to run in that team). The Be OS has no concept of
fork() or exec(), they're instead implemented
in our Posix library.
System calls from a user program are made in the traditional
UNIX way of trapping into the kernel and then jumping to the
real code. Since the only way a user program can get into
the kernel is via a trap, our kernel is more robust because
an errant or malicious program can't just jump into random
kernel code and cause damage. The overhead to make a simple
system call is quite low, coming in at just over three
microseconds.
The mechanism for IPC in UNIX is traditionally through a
pipe or the System V shared memory and message routines. The
main mechanism for IPC under the Be OS is through named
message ports that have a queue for incoming messages. Be
message ports are unidirectional, and the incoming queue is
limited in size to prevent run-away processes from gobbling
all of system memory. We also support shared memory through
the create_area() and clone_area() system
calls, which allow one process to create an area and another
to obtain a mapping to that area. The higher-level library
code uses both message ports and shared memory to enable
communication between threads in different teams.
Another area that the Be OS differs significantly from the
traditional UNIX model is in how it handles files. The Be OS
attempts to avoid the issue of handling path names of
arbitrary length by presenting a concept of a file "ref". A
file ref is a unique reference to a file on a given volume
and is independent of the file's path. This avoids dealing
with arbitrarily long pathname and eliminates hard- coded
constants, such as PATH_MAX, that limit how long a file path
can be. Of course the standard C library and Posix server
cooperate to provide a standard pathname-style interface, so
existing code doesn't have to be aware of this.
There are other differences between the Be OS and UNIX, but
these are a few of the highlights. The Be OS is different
from UNIX, but that doesn't imply incompatibility, since our
Posix library works to cover the differences. The features
that the Be OS does provide offer a rich base for building
new applications that use multithreading and messaging in a
real SMP environment.
There are still many areas we're working on in the Be OS,
and we plan on enhancing it significantly over time. Release
1.1d7 will contain a number of enhancements and we'll
continue to improve the Be OS as we get more feedback from
developers as to what they need to write even better, more
sophisticated applications.
MARKET OUTLOOK: Who the
Heck Wants a New Platform?
By Mark Gonzales
Here, at Be, we talk to people every day about the Be System
-- the BeBox and the Be OS. These people include developers,
customers, press, and resellers, all interested in the
potential of the Be System. They have plenty of questions,
and we provide lots of answers. But simmering just below the
surface is always one question: "Are you guys crazy? Who the
heck wants a new computing platform?"
Well, let's get this one out in the open. Conventional
wisdom in the computing industry says that the world has
moved towards a few standards on a limited number platforms
(read: one). No company can stand against the tide, it's a
better strategy to ride with it -- and that means being 100
percent compatible. To suggest something different is to
create astonishment in your audience. Be was called
"renegade" in one recent article.
If the question is "Who wants a new computing platform?",
I'll concede right now that conventional wisdom is
absolutely correct. The answer to this question is "No one
wants a new platform." But before you e-mail me in agreement
or disgust
(markg@be.com),
please allow me to place another thought on the table: "Is
this the right question to be asking?"
Jean-Louis is fond of saying that the source of product gold
mines is user frustration. And frustration with today's
personal computer systems is growing -- among both users and
developers. This isn't surprising since current computer
architectures are well over a decade old, and many of the
capabilities users and developers are demanding today simply
weren't considered when these architectures were designed.
Further, many changes have occurred in the industry since
these legacy systems first appeared. A few examples:
- Users are increasingly driven by a desire to
integrate audio, video, 3-D graphics, and other media
types into final publishable works. In this area,
interactivity is the key performance factor -- you want
to see the results as you manipulate the data. This
requires high-bandwidth, real-time data throughput and
high computational speed.
- The Internet has burst onto the scene, increasing the
demand for communications capabilities in PCs. But more
important, the world-wide web has created a new avenue
for publishing interactive information. Combined with the
rise of the CD-ROM and Digital Video Disk (DVD), we're
witnessing the first large-scale change in publishing
capability since the printing press, involving a richer
data set than static text and images. This is increasing
the demand for computing power from those creating the
new published works -- power to design and serve
web-based material.
- When current architectures were designed,
microprocessors were expensive. The concept of one
processor per person was revolutionary. Thus the software
architectures were designed to make the best use of a
single processor in each system. This design philosophy
stands today, even though today's processors are
low-priced commodities.
- Collectively we've learned much about software design
in the last decade. Yet this knowledge cannot be fully
utilized within the legacy architectures. The result is a
"muting" effect -- much of the performance delivered by
new hardware is lost in the overhead inherent in the
legacy systems.
A growing class of personal computer users are reaching,
and often surpassing, the limits of today's personal
computer architectures. They're purchasing high-end PCs and
then equipping them with expensive high-performance graphics
cards, video I/O cards, accelerators, and other add-ons, in
order to achieve minimally acceptable performance. In
essence, these users are paying extra because of the
overhead created by legacy system design, while receiving
systems that are underpowered for the same reason. These
users would purchase high-end workstations if they could
afford them, but most can't. All of this leads to an
understandable level of frustration.
For an industry that has always applauded itself for
delivering ever-increasing processor power, it can be a
sobering conclusion: A growing portion of the personal
computing audience is consuming processing power faster than
it's being introduced into the market.
Dataquest estimates that 68 million personal computers will
be sold in 1996, in addition to roughly 1 million
workstations. And growing at 4.5 times the rate of the
overall market is a class of machines Dataquest calls "PC
workstations." These personal computers -- delivering
workstation performance at PC prices -- will be flying out
the door at a 6-million-unit annual click within the next
few years. This class of PC will have tremendous influence
on the entire computing market -- impacting workstation
design as well as the look and feel of the average PC.
What will a PC workstation look like? Without a doubt it
will implement a level of processing power above what can be
obtained by straight-lining today's uniprocessor designs.
Most analysts project that these machines will need to be
based on multiprocessor designs. The PC workstation will
innately handle high-bandwidth media types such as video and
audio and will be at home on the Internet. In order to
maximize the capability and performance of these systems,
many analysts concede that the overhead within today's
legacy operating systems and applications will need to be
reduced. Tomorrow's operating systems will be significantly
different than today's, whether the products have familiar
brand names or not. And these PC workstations will have
simplified, modern, object-oriented programming models
allowing developers to concentrate more on their value added
than on system interfaces.
Sound familiar? It should -- we've just described the Be
System.
Will these PC workstations quickly appear on every user's
desk? Not immediately. Let's be frank, content consumers --
including most in the office automation markets -- are not
the ones pushing the architectural limits of PCs today. It's
the content creators who are running into performance
barriers. And they're the users to whom performance matters
most -- where getting a job done in an hour rather than a
day translates directly into improvement in the bottom line.
How many of these people are out there? Literally millions.
We'll talk about some of these types of customers in future
Be Newsletters, but you can probably already guess some of
the potential groups.
Who wants a new platform? No one. But this is an
industry-oriented question. We must think in terms
customers' desires and expectations. Customers think in
terms of capabilities, of interactivity, of getting the job
done in the best way possible -- as fast as possible. It's
better to ask, "Are there customers whose desires currently
exceed the maximum potential of today's PCs?" and "Do these
customers want to wait for today's PCs to slowly evolve
these desired capabilities?" These questions are more
customer-focused, and the answers are more
straightforward.
Conventional wisdom states that no company can stand against
the tide. That tide is made up of customer desires and
expectations, not product specs. Where customers demand,
companies must go. We at Be don't intend to stand against
the tide, we plan to ride with it. Because riding with the
tide, conventional wisdom tells us, is the best
strategy.
BE DEVELOPER PROFILE:
Acorde Corporation
To some developers, the BeBox's business potential is
still kind of murky. That comes as no surprise for a
platform that's only a few months old. But other developers
look at the BeBox's multiprocessor architecture and the
features of its operating system and find it hard to
resist.
Acorde Corporation of Provo, Utah, is an example. Acorde is
working on Apple's CyberDog product -- a version of Open Doc
for the Internet. They're also interested in graphics and
presentation applications.
But when Acorde looks at the Mac and other operating
systems, they see the past. Indeed, Apple is promising
multitasking, multithreading operating systems with Copland
and Gershwin, but they're not here yet. Be has already
delivered.
"We know that the capabilities of the other machines out
there are being stretched," says Stephen Humphrey, Acorde's
president. "The Mac is in a transition to a more modern
operating system for the future. But Be gives us a chance to
prototype and work with a platform that has those
capabilities now. And if Be takes off as a multimedia
platform, then we'll be there early."
Humphrey says he and his two associates at Acorde are
considering a port of Open Doc to the Be OS, because the
current lack of a component architecture is a hole in Be's
offering. But he's also impressed with Be's multimedia and
Internet capabilities. He considers the Internet to be the
next great publishing platform, and he thinks that Be's
hefty graphics and sound capabilities will present a good
opportunity there.
Humphrey started Acorde last October after working on Open
Doc at Novell. He's committed to the idea of a "cottage"
company, where people can work together over the Internet
from remote locations.
"The key is that we want to design for the future, rather
than the past, and this is the only way to do it," he says.
"It's an opportunity to architect applications from the
ground up that take advantage of threading capabilities. The
more advantage we can get by being on a cutting-edge OS like
Be, the better off we'll be."
Humphrey says he's hoping he can start playing with a BeBox
in the near future and have a prototype product ready by the
end of the year. Then he'll look to see whether there's a
market for it.
The Electronic Newsletter
More and more people want to know about Be Inc. and the
BeBox. Every day we receive new subscription requests. We're
really glad to see this enthusiasm and interest. In order to
please everybody -- and because of our weekly schedule -- we
have had to find a way to save time, paper, and money and to
be more flexible in updating our subscription database.
Therefore, starting with the next issue, we're going to send
the weekly Be Newsletter electronically -- this issue is the
last paper one.
To subscribe to this newsletter, subscribe to one of the Be
mailing lists, as described on the
Be Mailing Lists page on
this web site.
(Remember, if you're already receiving this newsletter,
you're already on the mailing list. You don't need to
subscribe again.)
|
