On running data-intensive algorithms with intelligent SSD and host CPU

Abstract

A solid state device (SSD), which has the characteristics such as high IO bandwidth and low access latency, is drawing attention as a next-generation storage device. Even though SSD provides a high internal bandwidth, the performance bottleneck exists on the host interface of relatively low bandwidth in spite of the increased internal bandwidth of SSD. To overcome the performance bottleneck, the notion of intelligent SSD (iSSD) has been proposed. In iSSD, there are still problems in processing the algorithms of high complexity. In this paper, we address an effective collaboration of iSSD and host CPU in order to maximize the performance of data-intensive algorithms. Extensive experimental results show that our approach performs faster up to 2.43 times than a previous approach.

High-performance data mining with intelligent SSD

“…In this paper, we selected kmeans [23], PageRank [24], and SimRank [25] because they are very popular in many application domains and also are data-intensive. SDF graphs for these algorithms are already constructed by our previous work [4] and we adopt them for our evaluation. Now, we present their SDF graphs.…”

“…In iSSD, each channel has an FMC core and DRAM and can process data stored in its own cells independently of those in other channels. FMC cores are thus primarily computing resources to perform ISP [1,4]. However, the gem 5 simulator only enforces their cores to share the memory.…”

“…Researchers have long recognized that a good way to process the big data more efficiently is to reduce the volume of data transferred over storage to a host system by taking computations in storage rather than in host CPUs. This is the reason that the notion of intelligent SSD (iSSD) was proposed to give computing power to SSD [1][2][3][4]. The iSSD is equipped with a core in each channel, thereby making it possible to process data inside iSSD without transferring data to host.…”

“…GPU is composed of a larger number of cores organized in multiple sets, each of which processes the same group of instructions in parallel with cores in it [8]. Some existing studies have addressed applications by using a single computing resource [1,9] while some other methods exploit collaborations of two different resources (e.g., CPU & GPU or CPU & iSSD) for more efficient processing [4,10]. We should address how to exploit all three computing resources for efficient processing of data mining algorithms since computing resources have their own characteristics.…”

“…In this environment, scheduling is required to decide which core is going to run for which function at which time, aiming at processing target algorithms in an optimal way [4]. We need heterogeneous scheduling [11,12] for our collaborative environment because it has computing resources of different types.…”