Back to cloudsim : BoT's

So this is the sequence:

• UA(user agent) on behalf of the user submits BoT's to BA(Broker Agent)
• BA calls a Sheduler Agent (SA) to schedule.
• SA then calls RA(Resource agent) representing VM's and allocates jobs (cloudlets)
• RA returns output to UA (bcos it's simulation... nothing is returned actually...)

Simulating BoT's (Bag-of-Tasks)

Size : 100 (Initially...)

No. of BoT's : 10

So 1000 cloudlets.... I'm just using the LCG log and i'm taking 1000 cloudlets 'n using break to stop reading more tasks from the log after 1000.

Next I wrote a class called BoT

Parameters

• no. of tasks
• submit time
• size of Instructions [NO_OF_TASKS]
• numPes [NO_OF_TASKS]
• sizeRatio

(there are other parameters such as ordering of tasks and the type of distribution which i've not included as of now...)

Methods

• BoT(int size) // constructor
• setSubmitTime(double s)
• double getSubmitTime()
• addTask(int, long, int)
• double calcSizeRatio()

After simulation, I've calculated the actual makespan of each Bot which is the largest finish time subtracted from the submission time and ideal makespan.

The ideal makespan is based on the concept that in theory, cloud computing has infinite computing resources... So, ideal makespan is the time taken for the largest task in a BoT to execute in the fastest machine available, I've given the formula as

max(sizeof task in BoTi) in million instructions / max (speed of resource j) in MIPS

These are the simulation results for the test run with the default time shared allocation policy in cloudsim:

Size of BoTs:	100		Speed of the Fastest VM :			1000 MIPS
BoT	Size Ratio	Submission Time	Max Finish Time	Ideal Makespan	Actual Makespan
1	2	19.6	76	20	57
2	2	436	496	20	59
3	3	971	1026	17	54
4	1	1398	1416	12	18
5	13	2039	2115	48	76
6	3	3010	3026	14	16
7	5	3740	3759	13	19
8	3	4053	4060	1	7
9	2	4314	4322	4	7
10	2	4918	4950	21	32

So now i've gotta implement my own scheduling algorithm for each BoT...