Currently this script only supports resource constrained scheduling
- Python
- lp-solve
On ubuntu, lp-solve can be installed as follows -
sudo apt-get install lp-solve
As of now it can only be used on linux systems (tested on ubuntu 16.04) as some linux specific commands has been used
A sample operation control graph has been provided in the repository. To see the script in action, please follow the following steps
- Clone this repo
- Open the cloned repo in terminal
- Run
python scheduler.py - You will be provided with a schedule on terminal, also output files for both the linear programming model and the output will be created in the same directory
The script requires the operation control graph as an input. This graph can be provided as a json input. That is a json file containing the definition of the graph along with some other parameters has to be present in the same directory as the source code. Currently a sample graph called ocg.json is already present in the repo. This file can be editted to provide a different graph input to solve and schedule.
Given below is a sample graph in json:
{
"total_ops": 8,
"csteps_estimate": 6,
"graph": {
"0": [1, 2, 3],
"1": [4],
"2": [5, 6],
"3": [7],
"4": [5],
"5": [6],
"6": [7]
},
"op_types_number": "4",
"op_types": {
"0": [0, 7],
"1": [1, 2, 6],
"2": [4, 5],
"3": [3]
},
"op_instances": {
"0": 1,
"1": 1,
"2": 1,
"3": 1
},
"var_type": "bin"
}
The above json file corresponds to the following graph
- The json must define total number of nodes, that is operations including the NOP operations
- Next it must define the estimated csteps or the desired csteps.
- Then the graph. Here each node/operation is given a number. The list corresponding to a given operation number signifies the operations that are dependent on the given operation. For example, in the above json file, the list corresponding to 2 contains 5 and 6 which means, 5 and 6 are dependent on 2. This can be verified from the picture where it can be seen that o5(/) and o6(+) depends on o2 (-). In this way the graph is created. This is nuothing but adjajency list represention of the OCG. In the graph declaration, NOP steps must be present.
- Next we need to mention the number of operator(resource) type available. Here it is 4, namely - multiplication(division), addition(substraction), lesser than and NOP (which is a dummy operator).
- Next we need to mention what are the operations present under each operator/resource type. For each operator type we need to provide a list of operations. For example, the list for 1 contains 1, 2 and 6. That is operations 1, 2 and 6 are of type 1. From the picture it can be verified that 1, 2 and 6 are either addition or substraction operation which can be considered as one, since for them we use the same resource (adder/substractor module)
- Finally we need to mention the variable type of our ILP problem. In this case it is bin(binary).
Finally we are done. On running scheduler.py we will be presented with the following output
Model generated and stored in lp_model.txt
Running lp_solver on lp_model.txt...
Processing output file generated by lp_solver...
Operation 7 should be scheduled on 5 cstep
Operation 0 should be scheduled on 0 cstep
Operation 1 should be scheduled on 1 cstep
Operation 2 should be scheduled on 2 cstep
Operation 3 should be scheduled on 1 cstep
Operation 4 should be scheduled on 2 cstep
Operation 5 should be scheduled on 3 cstep
Operation 6 should be scheduled on 4 cstep
Processing completed
Outpput stored in output.txt
As mentioned, the ILP model will be generated and stored as lp_model.txt and output (schedule) will be stored as output.txt in the same directory
- Currently it is assumed that number of resources available for all resource types is 1.
- It is also assumed that propagation delay for all operations is 1.
These will be removed in next iteration.