Recent graphic processing units (GPUs) often come with large on-board physical memory to accelerate diverse parallel program executions on big datasets with regular access patterns, including machine learning (ML) and data mining (DM). Such a GPU may underutilize its physical memory during lengthy ML model training or DM, making it possible to lend otherwise unused GPU memory to applications executed concurrently on the host machine. This work explores an effective approach that lets memory-intensive applications run on the host machine CPU with its memory expanded dynamically onto available GPU on-board DRAM, called GPU-assisted memory expansion (GAME). Targeting computer systems equipped with the recent GPUs, our GAME approach permits speedy executions on CPU with large memory footprints by harvesting unused GPU on-board memory on-demand for swapping, far surpassing competitive GPU executions. Implemented in user space, our GAME prototype lets GPU memory house swapped-out memory pages transparently, without code modifications for high usability and portability. The evaluation of NAS-NPB benchmark applications demonstrates that GAME expedites monotasking (or multitasking) executions considerably by up to 2.1× (or 3.1×), when memory footprints exceed the CPU DRAM size and an equipped GPU has unused VDRAM available for swapping use.

This paper deals with decentralized, QoS-aware middleware for checkpointing arrangement in Mobile Grid (MoG) computing systems. Checkpointing is more crucial in MoG systems than in their conventional wired counterparts due to host mobility, dynamicity, less reliable wireless links, frequent disconnections, and variations in mobile systems. We've determined the globally optimal checkpoint arrangement to be NP-complete and so consider Reliability Driven (ReD) middleware, employing decentralized QoS-aware heuristics, to construct superior checkpointing arrangements efficiently. With ReD, an MH (mobile host) simply sends its checkpointed data to one selected neighboring MH, and also serves as a stable point of storage for checkpointed data received from a single approved neighboring MH. ReD works to maximize the probability of checkpointed data recovery during job execution, increasing the likelihood that a distributed application, executed on the MoG, completes without sustaining an unrecoverable failure. It allows collaborative services to be offered practically and autonomously by the MoG. Simulations and actual testbed implementation show ReD's favorable recovery probabilities with respect to Random Checkpointing Arrangement (RCA) middleware, a QoS-blind comparison protocol producing random arbitrary checkpointing arrangements.

This paper deals with a novel, distributed, QoS-aware, peer-to-peer checkpointing arrangement component for Mobile Grid (MoG) computing systems middleware. Checkpointing is more crucial in MoG systems than in their wired counterparts due to node mobility and less reliable wireless links resulting in frequent and dynamic connections and disconnections. Having determined the globally optimal checkpoint arrangement to be NP-complete, we consider ReD, our Reliability Driven (ReD) protocol, employing QoS-aware heurisitcs, for constucting superior peer-to-peer checkpointing arrangements efficiently.