| tl;dr version: Computers and disks are a lot cheaper now. Basically the article boils down to this, what counted as a 'cluster' in 1998 is a single system in 2008, what used to take hundreds of disk drives to store, you can store on 1 today. Not particularly deep, but useful to think about from time to time. There is a quote, perhaps apocryphal, which says "There are two ways to solve a problem that would take 1000 computers 10 years to solve. One is to buy a 1000 computers and start crunching the numbers, the other is party for 9 years, use as much of the money as you need to buy the best computer you can at the end of the 9th year, and compute the answer in one day." The idea that computers get more powerful every year, and that in 10 years they will be more than 1000x more powerful than the ones you would have started with so one can solve the same problem. Of course they haven't been getting as powerful as quickly as they once were, but the amount of data you can store per disk has continued to outperform. The point is that if you are designing for the long haul (say 10 yrs from now) you can probably assume a much more powerful compute base and a lot more data storage. |
What he's saying is that the existence of the cloud and library advances such as MapReduce and APIs mean that the bar is lowered, when writing new software, to an extent it's hard even to comprehend.
Every time I get a module from CPAN I still get a shiver down my spine, remembering trying to do new and interesting things in the 80's and early 90's and every single time ending up trying to build a lathe to build a grinder to grind a chisel to hack out my reinvented wheel.