Hacker News new | past | comments | ask | show | jobs | submit login

Sure. It's trivial to arbitrarily reduce the probability of this one issue by throwing N times the hardware at it.

Now you have to deal with the problems of:

1) Buying, powering, managing, etc N times the hardware. When you buy hardware by the warehouse, this will be a big bill :-) 2) Since you are now increasing your network traffic, power, etc. by at least N times, it is very likely you will hit another problem that produces latency spikes.

Or, as Chris Colohan likes to say: "If you think solving problems is massively parallel computing systems is easy, please come join my team" :-)

I'm not saying that its easy, but the problem formulation in the article is simplified enough to be solved by this method. So I guess that my major objection is how the problem is stated.

Also in practice you don't really need N times the hardware. You pick out the queries with the worst jitter and either solve it or install two or three of these services, and you are good to go.

To summarize, I agree with you in principle, but would like to add that the method still is sound, but perhaps not in its most extreme form.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact