Projects / Programmes source: ARIS

intelligent GRID routing and scheduling [iGRIDras]

Research activity

Code Science Field Subfield
2.07.03  Engineering sciences and technologies  Computer science and informatics  Programming technologies - software 

Code Science Field
P170  Natural sciences and mathematics  Computer science, numerical analysis, systems, control 
Grid systems, artificial intelligence, machine learning, optimization, scheduling, routing
Evaluation (rules)
source: COBISS
Researchers (1)
no. Code Name and surname Research area Role Period No. of publicationsNo. of publications
1.  21305  PhD Daniel Vladušič  Computer science and informatics  Head  2007 - 2008  47 
Organisations (1)
no. Code Research organisation City Registration number No. of publicationsNo. of publications
1.  2012  XLAB software development and consulting Ltd.  Ljubljana  1639714  370 
Scientific and technological contributions of the post doctoral project are:  an intelligent and adaptive router and scheduler, based on artificial intelligence i.e., machine learning algorithms  a dynamic adaption algorithm for router and scheduler with properties of self adaption to the given environment.  implementation of the algorithm in the Linux kernel – integration into the XtreemOS Grid operating system. Scheduling and routing are two main research areas in the Grid. The overview of algorithms from both research areas shows that only simple (i.e., rather straightforward) algorithms are used for both: problems of routing and scheduling. Moreover, these algorithms are usually static in the sense of their flexibility and extensibility. While an suboptimal, but simple algorithm, with precisely defined input and output is still useful, the real obstacle to real-world use is the users inability to extend it with custom features, required by different uses of the system (e.g., policy for different classes of users). Such an algorithm would be far more useful if it supported different plugins that implement different profiles for different users, machines, etc. An additional problem lies in variety of Grid frameworks as each and every one of them supports only its own scheduling and routing algorithm – if the user wants to change it, he has to re-implement it. On the other hand, some of the researchers have recognized the variety of jobs run on the Grid systems and have suggested automatic learning of schedulers and routers (e.g., Boyan & Littman, 1994). These machine learning supported systems are adaptable, both in the sense of incorporating different criteria for estimating success of the router or scheduler and also in the sense of adapting to the underlying (physical) networks of computers and jobs. Our proposal is to extend these approaches with different artificial intelligence (machine learning) approaches, as the current approaches are rather simple. The approaches we’re planning to use range from fairly simple (KNN, LWR) to more advanced methods (e.g., regression and qualitative trees, neural networks). To ensure real-world usability, the extension of the current approaches on the administrative side would be incorporated. Our contribution would be to devise a constraint language, which could be used to define the allowed use of resources. The criteria implemented with this constraint language would be arbitrary, ranging from user credit to his importance in national sense. Finally, our proposal will be implemented in the form of a plug in for some of the major Grid frameworks. The (extensible) constraint language is architecture independent and would thus be seamlessly used and honored in learning and optimization process of our plug in. Our estimate of the applicative value of the project is very high. The trends in the literature and also in industry show an immense development of this field, as researchers estimate that there is still a lot to be gained with the current Grid architectures and solutions. Their estimate is that more intelligent and adaptive schedulers and routers are needed in order to get a bigger investment return with current technologies.
Significance for science
The results of the applicative project iGRIDras enable the use of the artificial intelligence methods in the GRID systems on the level of scheduling and routing. Despite such use of the artificial intelligence and machine learning methods is already known, the benefit of the iGRIDras project is in the machine learning methods used and the way they were used. Namely, within iGRIDras project the machine learning were used in the black-box manner, leaving the GRID administrator no parameters to modify or tune. The crucial result of the project is thus the research of the routing and scheduling problem with the aforementioned approach along with results which show the benefits of such approach. The project has shown the applicability of the artificial intelligence methods to the routing and scheduling problem in the GRID systems where we focused mainly on practical use in the sense of simple use for the system administrator. We emphasize that the approach was intentionally generic (e.g., minimum set of attributes used) which already yielded good results. We can conclude that tailoring of the approach to the target system should yield even better results. Still, this hypothesis was not confirmed within the iGRIDras project, as it was out of the project's scope. The developed framework is building on freely available PeerSim and GridSim frameworks, which ensures that the major frameworks from the field are covered in our work. Furthermore, further modifications arising from new versions of the aforementioned frameworks can be easily incorporated. This fact ensures the use of the iGRIDras framework in future developments in the field.
Significance for the country
Due to the specifities of Slovenia its future economic development depends mostly on research and development. Acquisition of new knowledge and expertise directly applicable to the XLAB's products and our many international projects, is thus extremely important, as it improves the company's business competitiveness as well as its academic level. It should be stressed that XLAB's specific internal structure makes it strongly depend on this kind of research and discoveries gained through the research. XLAB urgently needs quickly and easily applicable knowledge to expand the product line in Grid systems, SOA architectures and distributed systems in general. Discoveries in the area of intelligent routing and scheduling will be applied to improve the load balancing strategies in the ISL line of products as well as in various EU projects (XtreemOS, SLA@SOI etc).
Most important scientific results Final report, complete report on dLib.si
Most important socioeconomically and culturally relevant results Final report, complete report on dLib.si
Views history