Out of Core MapReduce for Large Data Sets OMR is a single machine MapReduce system to efficiently processes very large datasets that do not fit in main memory. It actively minimizes the amount of data to be read/written to/from disk via type-aware meta-data optimizations and on-the-fly aggregation. With minimized random disk I/O, OMR provides linear scaling with growing datasets, and significantly outperforms state of the art MapReduce systems like Hadoop and Metis. System details and performance results can be found in the ISMM'18 paper.
Following compilers/libraries are required:
- C++11 Compiler.
- Protocol Buffers
apt-get install libprotobuf-dev protobuf-compiler libgoogle-perftools-dev
To compile, go to the specific application directory and run:
make
For example, to compile word count application, run:
cd examples/wordcount
make
Note that above will produce two binaries:
wordcount.bin
(which uses variable sized records)wordcount-one.bin
(which uses fix-sized records)
You may need to raise the limit of maximum open file descriptors. You can do so by running:
ulimit -n 1048576
To run the application, you need to give the directory containing input files, number of mapper and reducer threads, batch size (which indicates number of key-value pairs that can be held by a thread in map phase), and kitems (which indicates number of key-value pairs that should be read by a thread from intermediate files during reduce phase):
<application_name> <folderpath> <nmappers> <nreducers> <batchsize> <kitems>
For example, to run word count with fixed-sized records:
./wordcount-one.bin ../../inputs/ 8 8 10000000 10000000
You can use other applications by replacing the binary name as follows:
./wordcount-one.bin ../../inputs/ 8 8 10000000 10000000
./wordcount.bin ../../inputs/ 8 8 10000000 10000000
./invertedindex-one.bin ../../inputs/ 8 8 10000000 10000000
./invertedindex.bin ../../inputs/ 8 8 10000000 10000000
./degreecount-one.bin ../../inputs/ 8 8 10000000 10000000
./degreecount.bin ../../inputs/ 8 8 10000000 10000000
./adjacencylist.bin ../../inputs/ 8 8 10000000 10000000
Gurneet Kaur, Keval Vora, Sai Charan Koduru, and Rajiv Gupta. OMR: Out-of-Core MapReduce for Large Data Sets. International Symposium on Memory Management, 12 pages, Philadelphia, Pennsylvania, June 2018.
To cite OMR, you can use the following BibTeX entry:
@inproceedings{Kaur:2018:OOM:3210563.3210568,
author = {Kaur, Gurneet and Vora, Keval and Koduru, Sai Charan and Gupta, Rajiv},
title = {OMR: Out-of-core MapReduce for Large Data Sets},
booktitle = {Proceedings of the 2018 ACM SIGPLAN International Symposium on Memory Management},
series = {ISMM 2018},
year = {2018},
isbn = {978-1-4503-5801-9},
location = {Philadelphia, PA, USA},
pages = {71--83},
numpages = {13},
url = {http://doi.acm.org/10.1145/3210563.3210568},
doi = {10.1145/3210563.3210568},
acmid = {3210568},
publisher = {ACM},
address = {New York, NY, USA},
}