Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Map - Split a problem up and distribute to workers

Reduce - Gather and aggregate/collate the results from the workers

That's it. That's the big thing to understand. The fact that everyone keeps looking for some new way to describe MapReduce I think adds to the mystique and the belief that it is somehow profoundly complex.

Aside from the many typos, I don't see how this is useful. MapReduce is an absurdly simple concept. Once you litter any explanation with analogies or scenarios, you distract the comprehension of the reader.

Am I to think about parsing text on a blog? Bad CEOs? Google and programs? What?



I think the article is just fine. People learn things differently. For some, a "real world" example can help drive things home.


I suppose, but there are better ways people can spend their time. This article is rife with misspellings and typos, which essentially shows a disrespect for readers. Secondly it's 1300+ words split into 6 "chapters" to say, poorly, what can be said in two sentences. I would not recommend it, especially on a site like HN (which purports to lean towards a higher standard).

In fact as a more general rule I would say that when people try to explain something, it is lazy, almost always confusing -- and usually an indication of the writer's own ignorance -- when said explanation resorts to contrived scenarios or analogies.

It's like if the UPS shipper had to deliver an elephant and twelve penguins, one of which suffered from gastroenteritis. By driving the truck using the hybrid energy recovery system, just how much conflict-gas that was fueled by the death of thousands of soldiers would a Catholic priest yield?


This makes sense to me than the article did and I had no prior understanding of it, though some people might also like real world examples.

I feel like seeing the essential (as you've done) and then finding examples that are specific implementations is how I learn best.


Not only is the algorithm described in the article complex for the simple task, but it is also incorrect. It has a single bottleneck: the grouper. It is interesting that I've been downmodded for providing a simpler but correct algorithm. (I don't know too much about MapReduce (almost nothing), but I like designing distributed algorithms for fun.)


He did a decent job of explaining how the different parts of the process relate to each other when it comes to the real world situation of parallelizing MR. The shuffle (grouper) actually happens per reducer in many implementations using hash-bucketed outputs from the mappers, but that's an optimization.




Consider applying for YC's Summer 2026 batch! Applications are open till May 4

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

Search: