Author: Michael Sevilla
Date: 6-22-2013
Institution: UC Santa Cruz
Email: [email protected]
HadooPhoenix is a collection of tools that implement scale-out algorithms on a scale-up system. The goal of this project is to compare the two most popular scaling architectures. The scale-out algorithms are modeled off the Hadoop implementatations and are implemented using Phoenix++. HadooPhoenix also includes sequential, 'dumbest thing possible' implementations. Using these two techniques, we cover a wide range of implementations, ensuring that we compare scaling architectures, not the applications themselves. A better introduction and references for all the tools and techniques we used can be found in the msevilla_masters_v0.pdf document in the ./docs folder of this distribution.
Files
*** Note - 'sop' stands for 'scale-out property' ***
- ./README.md: this file
- ./docs: supporting documentation with motivation, results, and references
- ./graphs: scripts that prepare output for gnuplot
- ./scripts: scripts for running scalability tests
- ./sop1_parallelism: uses Phoenix++ to achieve scale-out parallelism on scale-up
- ./sop2_fault-tolerance: uses DMTCP to achieve fault tolerance with checkpointing
- ./sop3_portability: library and shell script that queries hardware for optimal configurations
- ./sop4_storage: script that benchmarks the HDFS transfer speed (-copyToLocal)
We have reproduced the Phoenix++, DMTCP, and Phoenix2 source code so that the benchmarks for sop1_parallelism, sop2_fault-tolerance, sop3_portability, respectively, can be run as is.
End file