Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
A Library of Parallel Algorithms (cmu.edu)
212 points by federicoponzi on Nov 28, 2017 | hide | past | favorite | 20 comments


Back in around 2006 I was doing GPGPU regex and string matching and decided to google around for parallel string matching algorithms. I spent a good chunk of time scrolling around in the Vishkin's algorithm implementation and wondering why it all felt so familiar; it was only when I scrolled back to the top of the file that I realized it was familiar because I wrote it.

A fine example of why you should comment your code, as the person who reads it in 10 years from now will be a stranger, even if that person is you.


> A fine example of why you should comment your code, as the person who reads it in 10 years from now will be a stranger, even if that person is you.

For what it's worth, there is at least a comment with your name in it. Maybe I'm a narcissist, but those tend to jump out at me. ;- )


There is also the CMU free draft book on Parallel algorithms design http://www.parallel-algorithms-book.com/


This is great! I'm really inspired by the work CMU does on parallelism. They have a parallel programming course[0] that results in some really impressive final projects [1]. I also hear that CMU takes a "parallel-first" approach to teaching algorithms and data structures, but I have not been able to find out exactly how they do that.

[0]: http://15418.courses.cs.cmu.edu/fall2017/

[1]: http://15418.courses.cs.cmu.edu/fall2016/article/13


There's also 15-210 Parallel and Sequential Data Structures and Algorithms [1] which is what book GP referred to is developed for.

[1] https://www.cs.cmu.edu/~15210/


Animations link appears to be broken. After doing some way back machine snooping, looks like it hasn't been up since about 2000[0]. You can access the legacy java applet animations here[1]

[0] https://web.archive.org/web/20000611153404/http://web.scanda...

[1] http://www.cs.cmu.edu/~scandal/applets/


Thx, the images seem to be broken/nomore


These algorithms are written in NESL, which is perhaps the cleanest example of the oft-mentioned theory that parallel programming becomes simple in a functional setting. It really is remarkable how flexible it is. Unfortunately, I'm not sure NESL ever managed to run particularly fast in practice.


As someone who has done a lot of parallel programming, I really can't recommend Java 8 streams enough. They have a bit of overhead, however, they are incredibly easy to use and are extremely expressive.


I've done some tinkering with streams, and I was initially very excited. They do come with some frustrating limitations. Foremost in my mind: Concatenation of streams is fairly straightforward, but how do I perform a parallel join (zip?) of multiple streams?

In other words, given [1,2,3] and ["A","B","C"], how do I map over [[1,"A"],[2,"B"],[3,"C"]]?


AFAIK this isn't possible yet in a straightforward way, and it boils down to limitations of the type system (lack of tuples resp. value types in general), and the (understandable) refusal of many to do it in way that's hacky and potentially slow.

The discussion about this is actually quite old, for example: http://mail.openjdk.java.net/pipermail/lambda-libs-spec-obse...

IIRC that's a problem on bytecode level, that's why a (clean) solution still hasn't come up yet.


In the talk "Clojure Concurrency" [1], Rich Hickey demands that everyone in the room read "Java Concurrency in Practice" [2]. "It will scare the crap out of you."

[1] https://youtu.be/dGVqrGmwOAw?t=23m57s [2] https://www.amazon.com/Java-Concurrency-Practice-Brian-Goetz...


Java 8 released march 2014. Youtube video published december 2012. Book originally published 2006. What am I missing?


The fact Java 8 streams only address a tiny sliver of the concurrency problems addressed by JCIP.

In practice, Java 8 parallel streams are generally discouraged outside of raw number crunching, which is rarely where we need parallel speedup anyway.


Ah ok. I was confused by the reponse to slaymaker1907 who recommended Streams directly, as it didn't seem to address the Stream API. I guess many of the challenges of concurrent computing still applies here.

When else do we need parallel speedup?


raw number crunching doesn't parallel speedup(!)?


Didn't say that, I said that we rarely need parallelization for raw number crunching. Developers use parallelization for coordinating expensive IO way more often, and Java 8 streams solve almost nothing for that. Still need to know and worry about other Java concurrency primitives like volatiles, monitors, thread pools, etc. and all their pitfalls (race conditions, deadlocks, etc.).

You should still read JCIP, it is still highly relevant in Java-land. Java 8 streams solved almost nothing in terms of concurrency (in fact they probably created more problems than they solved...). You should still be scared of concurrency in general, and particularly in Java. And you should definitely consider a different language with more sane and modern concurrency primitives if you really need concurrency.


The parent might be saying that we already have some good techniques for speeding up raw number crunching in parallel, so it's not as desirable to add yet another.


What do that comment mean?


Now if only I used NESL for anything.




Consider applying for YC's Fall 2026 batch! Applications are open till July 27.

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

Search: