2 ms·
What Intel giveth, Microsoft taketh away? I'm not so sure. It's not as though Apache has gotten that much bigger. What's changed is the patterns of architectur
by jacques_chester 18y ago
What Intel giveth, Microsoft taketh away?
I'm not so sure. It's not as though Apache has gotten that much bigger. What's changed is the patterns of architecture; people getting lazy and then being stomped by traffic.
But you always have a kernel. You always have a web server. That part is progressively reduced by hardware advances.
- ars 18y agoApache can get bigger though. Right now it runs in threads, and in-process php parser (for example). Why? Because it's too slow otherwise. But get a faster machine, and people will get lazy. It's hard to code apache like that - much easier to have one process per request, and use one-run cgi scripts. Once that's fast enough, that's what people will code for. This is just an example, but I still think applications will manage to consume any extra CPU that comes their way. Example 2: Disk capacity, three years ago you would never have thought of a reason you'd need much storage, and now we have videos. The faster the desktops get, the bigger (and higher quality) the videos will get. That won't stop till a single 2 minute video will take 250MB. (At that point we'll need bigger monitors.) Example 3: Memory. One reason to use SQL cursors is so that you don't need to store the entire result set in memory at once. And used to be everyone programmed that way. But not anymore - memory is plentiful enough that queries are typically stored in their entirety in memory, and then walked through. They still use cursors to do that, but give the languages time to mature and you'll find it'll turn into random access arrays. It's the nature of people - they won't spend time to optimize something unless they need to. All your arguments assume usage (CPU, disk, bandwidth, memory), holds steady. Now assume they won't, assume they'll go up, and re-run your arguments.
- jacques_chester 18y agoGood point and well put. I'll give it some more thought, but my initial hunch is that the ratios of overhead will still converge, if for no other reason that software which doesn't expand as quickly (the kernel) gets factored out by the hardware-driven convergence on zero.