Benchmarking RHIPE vs the world.

[Marek Bejda]

Hello All, 

   We are trying to compare the performance of RHIPE, RHadoop,and Hadoop Streaming. Our first benchmarks are just typical wordcounts with varying number of mappers/reducers and different cases with large or lots of small files. Next week we'll be starting to compare KMeans runtimes. I am curious whether you know of any tests we can run where RHIPE shines or would yield greater performance vs the rest. Would you have any benchmarking/testing scripts that we could use or think would help us compare? 

  Anything that relieve us from reinventing the wheel would be very helpful :)

Thank you! 

Marek

[Saptarshi Guha]

I think the only timing we have is some massive linear regression. Ryan. do you know where it's at?

--

---
You received this message because you are subscribed to the Google Groups "rhipe" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rhipe+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ryan]

There is a technical report at Purdue on performance testing for logistic regression.  Let me check to see if it is publicly available.  It is a very interesting (and thorough) report and it shows how much some of the parameters matter. They are currently setting up a multi-factor designed experiment to study performance that I think will include different hardware settings in addition to parameter settings, which should be very interesting as I haven't seen something like that before - it's usually a random search, tweaking different parameters individually and following a path that gives incrementally better performance, without  much rigor.

BTW, I'd be very interested to see the results.