Yes, JNI should be good. Or you could go the hadoop pipes way - writing your MR apps in C++ and add the DLLs as dependencies distributed via distributed-cache.
On May 23, 2012, at 12:47 AM, jason Yang wrote:
> Currently, I'm trying to rewrite an algorithm into a parallel form. Since
> the algorithm depends on lots of third-party DLLs, I was wondering would I
> call the DLL written in C++ in the Hadoop-version MapReduce by using JNI?
> YANG, Lin