This project runs Simplex Method implementation, to solve a batch of LPs on a GPU.
System Configuration:
The project is tested on a 4 Core Intel Xeon CPU E5-1607 v4, 3.10 GHz, 63 GB RAM with Nvidia’s Tesla K40m GPU. However, it should run on any GPU.
For two set of LPs one on a randomly generated LPs and on a set of benchmarks from Netlib repository.

The LP benchmarks from NetLib repository has been chosen only when it satisfy our limitation on size. 
The platform of the project is 
OS is a 64bit Linux Ubuntu 14.04, we used the Editor as Nvidia's Nsight using the C/C++ CUDA implementation.

For eg to Run the Netlib benchmark number 7 in the paper with batch-size of 1000 LPs
using 10 streams   type the following command on the terminal (the script file BLPGSolver is executed which runs the executable in the folder bin)

 ./BLPGSolver 7 1 1000 10 1
 
 The name of the application is BLPGSolver to mean Batched LP Solver in GPU,
 1st Argument: indicate the benchmark number (option is 1 to 8 as presented in the paper)
 2nd Argument: indicate the averaging number
 3rd Argument: indicate the size of batch i.e., the number of LPs to be solved
 4th Argument: indicate the number of streams to be created, use 1 when batch-size is upto 100 otherwise 10
 5th Argument: can be 1 to decides to execute the Netlib benchmarks and 2 to decide to choose random LPs of dimension as supplied by the 1st Argument 




Dependency/Requirement:
1)  GLPK to be installed, we use version 4.45
2)  Boost Library (especially the boost_system and boost_timer )

(however the project also contains a pre-compiled libraries in our 64bit experimental setup, in the folder lib)
We observed that in some configuration some benchmarks does not give correct result, we suggest to use BLPGSolver-Debug in that case.

We also provide the source codes
1) BLPG-Solver : containing the actual source code required to generated the excutables in the folder bin (BLPGSolver and BLPGSolver-Debug)
2) CPLEXvsGLPK : containing the actual source code required to run a comparison between GLPK and IBM CPLEX. However, it requires installing CPLEX to be able to run this project. We have used the academic version of CPLEX.