Distributed job scheduling on computational Grids using multiple simultaneous requests

Exploring decentralized dynamic scheduling for grids and clouds using the community-aware scheduling algorithm

“…In general, metascheduling solutions are classified into three categories, namely the centralized, hierarchy, and decentralized schemes [18,10,15]. Specifically, the decentralized metascheduling scheme allows each node to own a metascheduler to receive job submissions originated by local users, and to assign such jobs to the local resource management system, i.e., local scheduler, of the node.…”

“…The K-Distributed and K-Dual Queue Models [15,21] propose a distributed scheduling algorithm which redundantly distributes jobs to different sites simultaneously, instead of only sending jobs to the most lightly loaded sites. The K-Distributed model enables each metascheduler of each site to distribute their jobs to the K least loaded sites, whereby such jobs will be scheduled by all K sites respectively.…”

“…Even concerning a centralized grid wherein each node maintains complete global information and is interconnected upon a static and stable network, the widely adopted backfilling series algorithms [13,14] can still lead to complex dynamic problems. Related research [15] has shown jobs requesting short processing time and few processors are likely to find ''holes'' easier in heavily loaded systems, which makes the prediction of future system performance as static even harder. In this case, a novel scheduling heuristic supporting dynamic rescheduling through time has shown its great importance, which is also covered in this paper.…”

Oferecimento de QoS para computação em nuvens por meio de metaescalonamento

“…No modelo centralizado, Figura 3.1, há uma Metaescalonador principal que mantêm informações de carga de trabalho sobre todas as organizações participantes da Nuvem (Subramani et al, 2002). Todos as requisições que são enviadas para a Nuvem são recebidas por um Metaescalonador que vai tomar a decisão de que organização irá executá-la.…”

“…Adaptado de (Subramani et al, 2002) No modelo distribuído, Figura 3.2, cada organização tem seu próprio Metaescalonador que periodicamente consulta os outros Metaescalonadores para obter informações de carga instantânea da Nuvem (Subramani et al, 2002). As requisições são enviadas…”