Allen,
Ning appears a bit beside himself, while at the same time making some very good points:
From: ning2m <ning2m@y...>
Date: Tue May 30, 2000 6:03pm
Subject: RE: QNX and explosive growth
Allen said "If Ning's assertion is true, then general-purpose OSes, like what you find on desktops and servers, will take over network equipment."
I hate to repeat myself but I have said that there is a huge difference between general OS and embedded OS and that the embedded OS is more appropriate for network equipment market (see my posting on 5/17/00).
The point I was making is that the market demand is driving many advanced features from general purpose OSes to embedded OSes (and RTOSes, which is a subclass of embedded OSes). For example, a pretty recent addition to VxWorks is the virtual memory, which is a common feature in general purpose OSes for years. WIND's next OS will have memory protection, which is again a common feature in general purpose OSes. So QNX has done what WIND is trying to do. I wish Allen would directly address any advanced features in QNX that he dislikes rather than attacking general-purpose OSes, which is NOT the point of discussion here.
Allen said "The issue is not whether QNX is the Holy Grail of OSes that can handle a desktop while also shrinking to a footprint..."
I didn't know QNX can handle desktop. If Allen knows an example where QNX is used as a desktop OS, please share with us. If not, please stop associating QNX with desktop OSes with innuendos.
Allen said "...ASICs' inflexibility has given rise to a new breed of intelligent network processors that most likely embody the future."
ASIC is anything but inflexible. In fact, ASIC is so flexible that it requires a lot of effort to work with it. The motivations for network processors are 1) to reduce the flexibility (thus the development time) by offering a inflexible canned solution to a subclass of network problems. 2) to reduce the development cost. 3) to reduce the production cost.
I think there is a niche for network processors (especially in the low performance, commodity network equipment market). However, I don't think network processors will be prevalent in the network equipment market, at least in the next few years. The reasons are:
1) Better development tools significantly reduced development time and cost. For example, to implement an algorithm in ASIC that used to take a hardware engineer 12 months to do in VHDL can now be done by a software engineer in 3 months in C.
2) You can now do prototyping, development and small-quantity production with FPGA and switch to ASIC only when you need to produce a large quantity of a relatively mature product. The FPGA design can be used directly with ASIC without additional development efforts.
3) ASIC offers much higher performance than network processors. I will discuss this in the next point.
Allen said "Moore's Law's loosing battle against the Law of the Telecosm..."
This is a very astute observation. However, his conclusion "...that speed will be a primary consideration in software solutions." is wrong. If you take the average networking software in the market and optimize it to the theoretical limit, you are getting just a few percentage gain in speed. Even if you took the worst piece of crap in the market and optimize it, you would not get more than 10 times (or 1000%) increase in speed. Therefore, optimizing the software is not a viable solution to bridge the gap between the Moore's Law and the Law of the Telecom.
So how do we solve this problem? I assert that the solution is ASIC.
I could substantiate my assertion with tedious technical discussions. But to keep it interesting, let me use an analogy. Let's look at our own brain. When we are first born, our brains are like unprogrammed FPGA. As we grow up, we are programmed to recognize faces (among many other things we do). Human neurons are operating at much slower speed than the gates (or a collection of transistors) in today's computers, yet we can recognize faces better and faster than the fastest computers. The secrete is in the specialization of a set of neurons for face recognition and the parallelism inherent in a web of large number of neurons. There are no CPUs in our head. Our brain is a collection of ASICs, each of which is programmed to perform a specific function.
So the nature is telling us that the path to high performance is not the ever increasing clock speed of transistors but ever expanding collection of ASICs.
Another interesting observation of ASIC is that the line between hardware and software is finally gone. There is no software instructions to be read into a machine and executed. The hardware is the embodiment of the software. The software is realized as the web of interconnections, just like the essence of us is realized as the web of interconnections of neurons.
There is a saying that when you hold a hammer in your hand, everything looks like a nail. Intel, being a CPU company, obvious would think every computing device needs a CPU. As you can see, this CPU centric view is having problems with the Law of the Telecom.
Allen said "Becoming the Microsoft of the post-desktop era is the only acceptable outcome for WIND, and it is attainable. This is a simple consequence economics in a competitive market with a zero marginal cost of production. There is no in between; its all or nothing."
Wow! Perhaps Allen is using hyperbole to make his point.
I think this view is too narrow and simplistic. As I said before, a competitive market is dynamic and chaotic. It's unlikely that we will see a repeat of PC market in the post PC era. There is a wide range of possible outcomes for WIND. Although I am generally bullish on WIND, I am unconformable to subscribe to such dogma. I think tunnel vision is dangerous in investing. Having healthy respect for the uncertainties in this dynamic market is only prudent.
Ning |
