4 ms·

"Oh, and by the way, RethinkDB is hiring right now to create the next generation of products that will fix all these problems with SQL." It's an interesting ar

by madair 17y ago

"Oh, and by the way, RethinkDB is hiring right now to create the next generation of products that will fix all these problems with SQL." It's an interesting article. But isn't he just saying, P=NP is a problem, so let's ban P=NP? Is this programmer newspeak?

coffeemug 17y ago
No. What I'm saying is that P=NP is a problem, so don't put O(2^N) algorithms into your production software. People seem to be offended by the fact that you can't run certain queries in realtime the moment you start talking about scale. It isn't my arbitrary definition, it's a fundamental limit in the physical universe.
- madair 17y ago
  I get it. I just wanted to point out that's what we're talking about, because it's also still a P=NP problem to even know if something is a P=NP problem, so all we can do, it would seem, is put up guards against the likely suspects. And that's all great. I'm trying not to be a negative nancy, but its statements like these that just seem so sloppy: "If this holds true, even if we can trivially parallelize each query, an exponential function (the amount of information) divided by a polynomial (number of machines) still dominates the logarithmic function we defined earlier as acceptable. This means that given modern trends, if a given query isn’t scalable vertically, it also isn’t scalable horizontally, which makes SQL fundamentally unscalable, period." Cut out the fluff and all that's saying is "Given modern trends, SQL is fundamentally unscalable, period." And why? Because anyone can write a bad query. And what's the answer? Complexity guards? Well, it certainly sounds good, more guards in SQL, or let's not call it SQL if there are organizational issues (i.e. non-modular design) and syntactical paradigms to improve. And lets insert some other [undescribed] computational theory to improve it, but the paradigms don't change the algorithmic reality, they can only hope to strategize better. Having said all that, I don't want to be misunderstood, it sounds like there may be some very interesting ideas behind all of this. It's hard to know in a case like this whether to comment on something, considering that much of the comment is negative. What motivates me is that in bullet points the article seems to imply something quite different than it sounds like when taken at face value. And I honestly appreciate push back if I'm just wrong about that. It's all about filters required to find the good stuff in all the the hyperbole that surrounds us.

by madair 17y ago

coffeemug 17y ago

No. What I'm saying is that P=NP is a problem, so don't put O(2^N) algorithms into your production software. People seem to be offended by the fact that you can't run certain queries in realtime the moment you start talking about scale. It isn't my arbitrary definition, it's a fundamental limit in the physical universe.

madair 17y ago
I get it. I just wanted to point out that's what we're talking about, because it's also still a P=NP problem to even know if something is a P=NP problem, so all we can do, it would seem, is put up guards against the likely suspects. And that's all great. I'm trying not to be a negative nancy, but its statements like these that just seem so sloppy: "If this holds true, even if we can trivially parallelize each query, an exponential function (the amount of information) divided by a polynomial (number of machines) still dominates the logarithmic function we defined earlier as acceptable. This means that given modern trends, if a given query isn’t scalable vertically, it also isn’t scalable horizontally, which makes SQL fundamentally unscalable, period." Cut out the fluff and all that's saying is "Given modern trends, SQL is fundamentally unscalable, period." And why? Because anyone can write a bad query. And what's the answer? Complexity guards? Well, it certainly sounds good, more guards in SQL, or let's not call it SQL if there are organizational issues (i.e. non-modular design) and syntactical paradigms to improve. And lets insert some other [undescribed] computational theory to improve it, but the paradigms don't change the algorithmic reality, they can only hope to strategize better. Having said all that, I don't want to be misunderstood, it sounds like there may be some very interesting ideas behind all of this. It's hard to know in a case like this whether to comment on something, considering that much of the comment is negative. What motivates me is that in bullet points the article seems to imply something quite different than it sounds like when taken at face value. And I honestly appreciate push back if I'm just wrong about that. It's all about filters required to find the good stuff in all the the hyperbole that surrounds us.