A Time Optimization Algorithm for Scheduling Bag-of-Task Applications in

Proportional Share Systems

Basically, there are various economy models that could be selected to apply into Grid system as the strategy of resource trading. However, since the characteristics of the Grid system are different from other computer systems, the auction becomes the most suitable model to be applied in the Grid system. Under the Auction model, the resource price depends on how the user views the value of the services. The user who bid the highest price earns the qualification to access the grid services [2].

The action is one of the important market mechanisms. Under this mechanism, tthere is no fixed price standard but depends on the user demand. The buyer and the seller who participate in the auction choose the strategies rationally to maximize their own expected values. The time optimization algorithm by Dr. Buyya uses the Auction model and combines with the proportional share system. It focuses on “bag-of-task applications” which means there is no communication application as the scheduling between the sub tasks. The major objective of the algorithm is to reduce the total run time to meet the user deadline. When all tasks are able to be finished within the deadline, trying to cut down budget could achieve better utility. Besides, the concept of adding VO of this algorithm offer the user and the resource the opportunity to integrate into the same virtual community to obtain better resources and better premiums. In other words, if users want to make use of resources of different VO to compute the jobs, they have to pay extra expenses.

The procedures of the time optimization algorithm by Dr. Buyya are introduced below.

Each VO has its own Grid Information Services (GIS) which is responsible for recording the information of resources in community. Before computing the jobs, the user broker asks for the information from GIS to process job scheduling. After choosing appropriate resource, the jobs and budget are conveyed to the resource broker to proceed to the auction and obtain the resource share. During each auction, the user broker adjusts bidding price according to the expected completed time of the jobs. After completing the jobs, the user broker calculates and pays the total expenses. If the user chooses recourse from different VO to computing jobs, he/she has to pay extra expenses. Therefore, the time optimization algorithm by Dr. Buyya could be divided into two categories, the Global and the Local policy. In the one hand, under the Global policy, the user broker searches usable resources from different VO in the grid system to complete the jobs. In other hand, under the Local policy, the user broker searches

resources from the same VO as the user to proceed computing. There are pros and cons in these two policies. There are more resources to be chosen under the Global policy, however, extra expenses have to be paid when selecting resources from other VO to compute the jobs.

As for the Local policy, there is no extra fee but fewer usable resources. Besides, if the user broker evaluates and finds out the jobs couldn’t be finished within deadline under the Local policy, he/she would search resources from other VO to compute and meantime has to pay more money because of the time pressure to complete the jobs [29].

The characteristics of this algorithm are the computing ability of using the proportional share system to allocate the resources. The standard of allocation is not the weight model as the usual but the price user offers. This approach better corresponds to the principle of economy model. If users need to complete the jobs earlier, they are willing to offer higher price to obtain more proportion of resources. Moreover, the user broker adjusts the resource price amount according to the user’s deadline and budget in each auction time. The major goal is to complete the all jobs within the deadline and the second is to reduce total costs.

The Time optimization algorithm by Dr. Buyya mentions that users could set time period P by themselves to bring their prices to resources elastically depending on the order and priority of the jobs. However, we couldn’t carry it out in the study and couldn’t prove if it really has utilities. We will address the improvement aimed on this point in the research. The second part of the research focus on bringing up the inquiry that if the user broker initialing the price is too conservative. Basing on its principle, the user broker would take the resource with worse computing utility as the initial price standard to prevent operating jobs after the deadline in the worst case. However, if the gap of the difference of the resource utility is too big comparing to other resources, the excessive initial price would exist. Under this condition, there must be some wasting situations. Aiming at the two cons mentioned above, this study

offers the improved algorithm called A Flexible Job Scheduling with Auction-based Proportional Share Model. The detail about the framework and the method are introduced in chapter 3.

In this chapter, we introduce the advantages of the Grid Economy and the GRACE framework at first. In 2.2, Economics Models are introduced based on the GRACE framework.

The four algorithms applied in Nimrod-G are illustrated in 2.3. These four algorithms focus on improving the user deadline or budget but none of other factors are considered. Therefore, L Kenli, T Xiaoyong, Z Zhaohuan and Dr. Buyya add the concept of VO in it to make the job scheduling more complete. Sai Rahul Reddy P. also comes up the AHP policy making the economy model algorithm more complicated with multidimensional and weight approaches.

We found out the auction model more appropriate to be applied into the Grid system as the resource management. The Time Optimization Algorithm by Dr. Buyya that combines the auction model and the proportional share model has the effect but is too conservative.

Therefore, on the basis of the research of Dr. Buyya, we design the more flexible auction system with proportional share model as the job scheduling strategy.