Performance Matters ... and why
Driving to the office this morning, I was thinking about an up-coming game changer. I won't divulge just now what game changer I was thinking of. But I am concerned that it will lack vital instrumentation. This thing is going to be hot, and it will ride a wave of investment. Some might say, "Performance? It's good enough! Measurement? Why bother?".
I did not previously care all that much about performance. It's not sexy. Performance measurement is statistics. Ewww. In fact, the sexier stuff, graphical interfaces and cloud-capable applets, tend to be horrible in terms of resource consumption. If you shine the glare of the performance spotlight on them, they don't look quite as flashy as they did when surrounded by smoke and mirrors. Up to now, it did not matter.
But performance (more accurately, our delay in attending to it) will come back to bite us. This is especially true when we have to run a task over and over. (Isn't that how we got into computers to begin with? We want the machine to carry the tedium.) It helps when the machine actually works. It helps when it's quick about the task we give it. The advent of smart phones and other embedded computing shows us that performance matters a lot; it becomes a part of basic reliability.
I came to respect the performance experts and to value the performance question. As a multi-platform guy, I always valued the different platforms for their unique strengths. But I found myself defensive of my favorite platform. It's expensive. (It's also not flashy ... at least not as flashy as a Mac or even most PCs.) The justification for this beast is in performance. And if you can't measure it ... well ... that's a whole nutha story. But we can measure it. And we did. And we demonstrated the value.
There will come a new player soon. (There are always newcomers.) It will slip into a niche of the industry with enough recognition to be welcomed. But it will not be mature. It's going to be cool, even glitzy. Those who embrace it will give it ample resources. It will be good enough.
Throwing money at the problem does not scale. Whether people or processors, you can't always escalate. Sometimes the "system" needs to be tuned. And tuning is not cooked like frog legs in the skillet. Tuning requires measurement. Adding more and more power is just not good enough.
I look forward to this new thing, and others. But like Ford's latest ride, I want it tested, measured, ... analyzed to pieces. First, I want to know that the air bags will deploy. But I also want to know the pinch at the pump. It matters.
-- R; <><