A performance evaluation of scientific I/O workloads on Flash-based SSDs

Abstract

Abstract-Flash-based solid state disks (SSDs) are an alternative form of storage device that promises to deliver higher performance than the traditional mechanically rotating hard drives. While SSDs have seen utilization in embedded, consumer, and server computer systems, there has been little understanding of its performance effects with scientific I/O workloads. This paper provides a trace driven performance evaluation of scientific I/O workloads on SSDs. We find that SSDs only provide modest performance gains over mechanical hard drives due to the writeintensive nature of many scientific workloads. Other workloads (like read-mostly web servers) would likely see much larger gains. Additionally, we observe that the concurrent I/O (when multiple parallel processes simultaneously access a single storage device) may significantly affect the SSD performance. However, such effects appear to be dependent on specific SSD implementation features and they are hard to predict in a general fashion. These results suggest that abundant cautions are needed when supporting high-performance scientific I/O workloads on Flashbased SSDs.

Understanding the Impact of Emerging Non-Volatile Memories on High-Performance, IO-Intensive Computing

“…These include studies of the basic performance properties of SSDs [8] as well as their system-level performance for particular domains such as scientific [28], [33], [2], data center/database applications [24], [23], [32], and E-business [21], [25]. Emerging technologies have also sparked interest in their usefulness in building so-called ExaScale systems [1].…”

DASH: a Recipe for a Flash-based Data Intensive Supercomputer

“…The speedup is substantial but not as good as expected, which could be explained by the mix of bandwidth and latency bound (sparse and dense) accesses in traversing the test graph. As previous works [14] observed, writeintensive nature of an application might also be the cause. …”

Incorporating Network RAM and Flash into Fast Backing Store for Clusters

“…There has also been much recent work that examines incorporating emerging solid-state storage into systems [12], [13], [14], [15], [16]. Different uses include incorporating flash into the memory hierarchy and using it as a fast out of core buffer cache in database management systems.…”

“…Some work has examined how flash can be used in high-performance computing. In [15], the authors compare scientific workload run times for different backing storage devices: two flash devices and disk. They find that for sequential workloads, the flash drives offer little improvement over disk, but for parallel workloads flash significantly outperforms disk, primarily due to increased opportunity for parallel I/O.…”