Accelerating dynamic graph analytics on GPUs

Abstract

As graph analytics often involves compute-intensive operations, GPUs have been extensively used to accelerate the processing. However, in many applications such as social networks, cyber security, and fraud detection, their representative graphs evolve frequently and one has to perform a rebuild of the graph structure on GPUs to incorporate the updates. Hence, rebuilding the graphs becomes the bottleneck of processing high-speed graph streams. In this paper, we propose a GPU-based dynamic graph storage scheme to support existing graph algorithms easily. Furthermore, we propose parallel update algorithms to support efficient stream updates so that the maintained graph is immediately available for high-speed analytic processing on GPUs. Our extensive experiments with three streaming applications on large-scale real and synthetic datasets demonstrate the superior performance of our proposed approach.

Packed Memory Arrays - Rewired

“…The ST are those determined in Section III. The UT are derived with ρ 1 = 0.08, ρ h = 0.3, τ h = 0.75 and τ 1 = 1, mimicking the configuration of previous work [10], [14], [30]. These are also the density thresholds employed in the rest of the experiments.…”

“…For consistency [10], this approach imposes a further constraint: 2 · ρ h ≤ τ h . For this strategy, common thresholds in existing applications [10], [14], [30] are ρ 1 ∼ 0.1, ρ h ∼ 0.3, τ h ∼ 0.75 and τ 1 ≥ 0.9. In the second approach, the capacity of the array is set to 2·N τ h +ρ h on resize.…”

“…• We present a new bulk loading algorithm. It is particularly suitable for the streaming scenario [30], [31], where the cardinality of the array is kept constant, and updates, featuring the same amount of insertions and deletions, are executed in batches at regular intervals. This paper is organised as follows: In Section II, we summarise PMAs and their properties.…”

Efficient Representative Subset Selection over Sliding Windows

“…Subsequently, for each checkpoint x i , H(x i ) processes v t and updates the candidates independently according to Lines 6-13 of Algorithm 1. After updating the candidates of H(x i ) for v t , it performs the buffer maintenance procedure as follows (Lines [25][26][27][28][29][30][31].…”

“…In many cases, data is generated rapidly and only available as a stream [11,18,19,30,39]. To address the requirement for summarizing such datasets in real-time, RSS over data streams [2,10,29,32] has been extensively studied in recent years.…”