4 ms·

I've studied Christensen's three main books; I find them very appealing but I'm not quite convinced by the component-vs-integrated idea. While an integrated ap

by 6ren 14y ago

I've studied Christensen's three main books; I find them very appealing but I'm not quite convinced by the component-vs-integrated idea. While an integrated approach can squeeze out more performance, does it predict industry structure? Here, it's true that the iPhone's integration greatly aids its performance; and performance is certainly needed in this market. But (IBM) PC's were never integrated... and performance was certainly in demand in that market (for years and years people would keep buying faster PCs). And, to compare them with Apple's Macs at that time (which were integrated, just like the iPhone), PC's often had better performance per buck than Macs. Christensen argues that this is because the performance enhancement was more due to particular components (especially the CPU), rather than their integration. Thus, those key components became the focus for improvement. So, the result of Intel improving CPUs (for example) was that the performance of the whole PC increased. This saves the theory... in fact, he turns it around, and points out that Intel CPUs were integrated internally in order to get that performance. While I do find this very interesting, I get the feeling of post hoc analysis in too many of his examples... and a theory that doesn't predict the future is just not quite compelling for me. :-) BTW: Christensen says that integrated and component approaches have different strengths, and which one wins depends on what that market needs. If the market is desperate for performance (and will pay a premium for it), then an integrated approach will be more popular and profitable. But if the market doesn't really care about performance, but instead wants something that is customizable, then a component approach will win. Markets change over time. Typically, some feature gets better until it's "good enough" (most people won't pay for more of it). He claims there's also an alternation between the levels of a system: if "integration" wins at one level, then "componentization" wins at the parts that are integrated e.g. PC's were not integrated, but their parts (CPUs, HDD etc) were integrated internally... but the parts used within them switched back to being not integrated.

joe_the_user 14y ago
I haven't read Christensen in particular but there's another factor which could distinguish PCs and Phones. PCs were in demand for more or less one quality, power (with screen resolution making maybe two). A phone is in demand for a multitude of qualities - size, power, batter life, screen resolution, ergonomics, etc. A PC that was twice the size, twice the power and half the price of its competitor could do well. Phone not so much. With a multitude of constraints to combine, it's not surprising that software hardware integration happens. Cars are integrated systems more than a hundred years after their invention whereas you can make do buying different components for your house.
- 6ren 14y ago
  First, I'll correct my terminology - it's not "componentized" but "modular". Both integration and modular approach have components, the distinction is that different modular components can be used; there are many components that fulfill the same interface in a modular architecture, whereas in an integrated architecture, the components are interdependent, so you can't just swap one out. It's similar to reusing modular code: you can have different implementations of a List, and they all work the same - and, since it's standardized interface, many different programs can reuse them. But in integrated code, if you change one thing, you have to change everything else. So you can't mix and match, and you can't reuse the code between projects. So "modularity" is really about the cleanly defined interfaces. Yes, I agree phones are more constrained. But for evaluating the theory, how can you tell how constrained it has to be, to count? For example, Carmack says that integrated CPUs and GPUs are about twice as fast than as separate modular components (so, for the same power components, an xbox360 can be twice as fast as a PC). And yet, even though graphics power is in demand, the PC market was (and still is) dominated by modular GPUs. Curiously, Christensen doesn't provide any statistical data across industries (but he did do this for his basic "disruption" idea). I think it's more that, if some quality is in demand, and if that quality can be delivered best by integration, then that might be how it happens. It's a softer theory. I guess the simple answer is that this idea of integrated vs. modular is just one factor of many, so it's not predictive alone. BTW: surprisingly, cars are not integrated according to Christensen - they are assembled from different components that can be mixed and matched (e.g. the same electrical system can be used in several different models; and different car, same engine; the same chassis can accommodate different versions of each component, because they have bolts in the same places, and use the same interfaces in other respects). This happened because people don't want faster cars (that integrated components are better at delivering); they want safer, more economical cars. But here I have doubts - it seems both those could be better achieved by integration (esp for efficiency).
  - joe_the_user 14y ago
    Hmm, Yes, I agree phones are more constrained. But for evaluating the theory, how can you tell how constrained it has to be, to count? That is a good question. It seems like one simple approach would be to just generalize the big three aspects of phones and cars: Space, Energy and Time. These are kind of the big, orthogonal constraints of our world. You can probably dig deeper and find some others but that's what occurs to me. People always want cars that are faster, more efficient and fit in a given form-factor. Same with phones. These matter less with desktop machines or say, houses. Laptops are somewhere in-between. I guess a low-end car might be somewhat less integrated than a higher end car. Etc...

by 6ren 14y ago

joe_the_user 14y ago

I haven't read Christensen in particular but there's another factor which could distinguish PCs and Phones. PCs were in demand for more or less one quality, power (with screen resolution making maybe two). A phone is in demand for a multitude of qualities - size, power, batter life, screen resolution, ergonomics, etc. A PC that was twice the size, twice the power and half the price of its competitor could do well. Phone not so much. With a multitude of constraints to combine, it's not surprising that software hardware integration happens. Cars are integrated systems more than a hundred years after their invention whereas you can make do buying different components for your house.

6ren 14y ago
First, I'll correct my terminology - it's not "componentized" but "modular". Both integration and modular approach have components, the distinction is that different modular components can be used; there are many components that fulfill the same interface in a modular architecture, whereas in an integrated architecture, the components are interdependent, so you can't just swap one out. It's similar to reusing modular code: you can have different implementations of a List, and they all work the same - and, since it's standardized interface, many different programs can reuse them. But in integrated code, if you change one thing, you have to change everything else. So you can't mix and match, and you can't reuse the code between projects. So "modularity" is really about the cleanly defined interfaces. Yes, I agree phones are more constrained. But for evaluating the theory, how can you tell how constrained it has to be, to count? For example, Carmack says that integrated CPUs and GPUs are about twice as fast than as separate modular components (so, for the same power components, an xbox360 can be twice as fast as a PC). And yet, even though graphics power is in demand, the PC market was (and still is) dominated by modular GPUs. Curiously, Christensen doesn't provide any statistical data across industries (but he did do this for his basic "disruption" idea). I think it's more that, if some quality is in demand, and if that quality can be delivered best by integration, then that might be how it happens. It's a softer theory. I guess the simple answer is that this idea of integrated vs. modular is just one factor of many, so it's not predictive alone. BTW: surprisingly, cars are not integrated according to Christensen - they are assembled from different components that can be mixed and matched (e.g. the same electrical system can be used in several different models; and different car, same engine; the same chassis can accommodate different versions of each component, because they have bolts in the same places, and use the same interfaces in other respects). This happened because people don't want faster cars (that integrated components are better at delivering); they want safer, more economical cars. But here I have doubts - it seems both those could be better achieved by integration (esp for efficiency).
- joe_the_user 14y ago
  Hmm, Yes, I agree phones are more constrained. But for evaluating the theory, how can you tell how constrained it has to be, to count? That is a good question. It seems like one simple approach would be to just generalize the big three aspects of phones and cars: Space, Energy and Time. These are kind of the big, orthogonal constraints of our world. You can probably dig deeper and find some others but that's what occurs to me. People always want cars that are faster, more efficient and fit in a given form-factor. Same with phones. These matter less with desktop machines or say, houses. Laptops are somewhere in-between. I guess a low-end car might be somewhat less integrated than a higher end car. Etc...