Improving read performance by isolating multiple queues in NVMe SSDs

Lee, Minkyeong; Kang, Dong Hyun; Eom, Young Ik

doi:10.1145/3022227.3022262

Cited by 22 publications

(10 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, Caulfield et al [19] proposed to bypass the BIO layer and implemented a separate driver, as a way to increase performance. Lee et al [9] presented a novel queue isolation scheme, considering the write interference and increasing the read performance in heavy read workloads. Bjørling et al [11] redesigned the BIO layer in order to enable scalability and exploit the spare hardware capabilities by implemented multi-queue support.…”

Section: Related Workmentioning

confidence: 99%

“…Furthermore, the submission/completion scheme of NVMe requests is configurable and allows multiple cores to share one completion queue, leading to more efficient memory utilization [8]. Although the default NVMe architecture includes up to 64K pairs of submission and completion queues, several research papers propose isolation schemes of write and read queues in order to improve performance [9].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

FastPath: Towards Wire-speed NVMe SSDs

Stratikopoulos¹,

Kotselidis²,

Goodacre³

et al. 2018

EasyChair Preprints

View full text Add to dashboard Cite

Abstract-The constant growth of data and its importance to drive Machine Learning and Big Data is pushing storage systems towards ever increasing I/O bandwidth and lower latency requirements. In recent years, the Non Volatile Memory Express (NVMe) standard has enabled SSD drives to deliver high I/O rates by allowing the storage to be connected directly via the fastest available interconnect to the processing chip. In parallel, the adoption of FPGAs in data centers is creating opportunities to accelerate various applications and/or Operating System (OS) operations. While, FPGAs in data centers have been connected via PCIe to mostly x86 servers, we have now also available heterogeneous SoCs with multi-cores and FPGAs integrated on the same die and connected by an onchip interconnect.In this paper, we present how to rethink and accelerate NVMe performance on heterogeneous SoC with integrated FPGAs providing a first research insight on the performance benefits of such an approach. We provide an analysis of the Linux block I/O layer and showcase the relationship between the system's performance and its I/O implementation. Consequently, we introduce an FPGA-based fast path which accelerates the access to the NVMe drive. Our comparative evaluation demonstrates that our FPGA-based FastPath achieves up to 71% lower latency and up to 5x higher I/O performance against the baseline system on a Zynq development board.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

FastPath: Towards Wire-speed NVMe SSDs

Stratikopoulos¹,

Kotselidis²,

Goodacre³

et al. 2018

EasyChair Preprints

View full text Add to dashboard Cite

show abstract

“…SCMs used with legacy disk based interface such as SATA/SAS were incapable of harnessing the throughput and inherent parallelism. However, recent advances such as faster PCIe bus technology (also known as NVMe Express) [48] [49], PCIe switches, Linux block layer redesign, etc. are enabling SCMs to provide higher performance.…”

Section: Need For Multiple Tiers and Automated Tieringmentioning

confidence: 99%

LDM: Lineage-Aware Data Management in Multi-tier Storage Systems

Mishra

Somani

2019

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

We design and develop LDM, a novel data management solution to cater the needs of applications exhibiting the lineage property, i.e. in which the current writes are future reads. In such a class of applications, slow writes significantly hurt the over-all performance of jobs, i.e. current writes determine the fate of next reads. We believe that in a large scale shared production cluster, the issues associated due to data management can be mitigated at a way higher layer in the hierarchy of the I/O path, even before requests to data access are made. Contrary to the current solutions to data management which are mostly reactive and/or based on heuristics, LDM is both deterministic and pro-active. We develop block-graphs, which enable LDM to capture the complete time-based data-task dependency associations, therefore use it to perform life-cycle management through tiering of data blocks. LDM amalgamates the information from the entire data center ecosystem, right from the application code, to file system mappings, the compute and storage devices topology, etc. to make oracle-like deterministic data management decisions. With trace-driven experiments, LDM is able to achieve 29-52% reduction in over-all data center workload execution time. Moreover, by deploying LDM with extensive pre-processing creates efficient data consumption pipelines, which also reduces write and read delays significantly.Abstract. We design and develop LDM, a novel data management solution to cater the needs of applications exhibiting the lineage property, i.e. in which the current writes are future reads. In such a class of applications, slow writes significantly hurt the over-all performance of jobs, i.e. current writes determine the fate of next reads. We believe that in a large scale shared production cluster, the issues associated due to data management can be mitigated at a way higher layer in the hierarchy of the I/O path, even before requests to data access are made. Contrary to the current solutions to data management which are mostly reactive and/or based on heuristics, LDM is both deterministic and pro-active. We develop block-graphs, which enable LDM to capture the complete time-based data-task dependency associations, therefore use it to perform life-cycle management through tiering of data blocks. LDM amalgamates the information from the entire data center ecosystem, right from the application code, to file system mappings, the compute and storage devices topology, etc. to make oracle-like deterministic data management decisions. With trace-driven experiments, LDM is able to achieve 29% to 52% reduction in over-all data center workload execution time. Moreover, by deploying LDM with extensive pre-processing creates efficient data consumption pipelines, which also reduces write and read delays significantly.

show abstract

“…Unlike the traditional SSDs with FTLs built-in to provide an abstracted block I/O interface, PASSD exposes a physical interface of NAND flash memory, and a FTL is located on a host system. It offers various advantages, such as predictable response time at the host system level, physically independent I/O path that can be used as a scheduling option (e.g., isolation) [26], and stack optimization with a VOLUME 4, 2016 file system. Therefore, the new class of SSD is expected to compensate for the shortcomings associated with legacy SSDs and provide better performances.…”

Section: Introductionmentioning

confidence: 99%

Improving Write Performance Through Reliable Asynchronous Operation in Physically-Addressable SSD

et al. 2020

View full text Add to dashboard Cite

Physically-addressable solid-state drives (PASSDs) are secondary storage devices that provide a physical address-based interface for a host system to directly control NAND flash memory. PASSDs overcome the shortcomings such as latency variability, resource under-utilization, and log-on-log that are associated with legacy SSDs. However, in some operating environments, the write response time significantly increases because the PASSD reports the completion of a host write command synchronously (i.e., write-through) owing to reliability problems. It contrasts asynchronous processing (i.e., write-back), which reports a completion immediately after data are received in a high-performance volatile memory subsequently used as a write buffer to conceal the operation time of NAND flash memory. Herein, we propose a new scheme that guarantees write reliability to enable a reliable asynchronous write operation in PASSD. It is designed to use a large-granularity mapping table for minimizing the memory requirements and performing internal operations at an idle time to avoid response delays. Results demonstrate that the proposed PASSD reduces the average write response time by up to 88% and guarantees reliability without performance degradation.

show abstract

Improving read performance by isolating multiple queues in NVMe SSDs

Cited by 22 publications

References 10 publications

FastPath: Towards Wire-speed NVMe SSDs

FastPath: Towards Wire-speed NVMe SSDs

LDM: Lineage-Aware Data Management in Multi-tier Storage Systems

Improving Write Performance Through Reliable Asynchronous Operation in Physically-Addressable SSD

Contact Info

Product

Resources

About