Initial Experience with 3D XPoint Main Memory

Liu, Jihang; Chen, Shimin

doi:10.1109/icdew.2019.00009

Cited by 9 publications

(9 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Optane DCPMM performance being lower than DDR, one may expect these NUMA effects to be negligible. Unfortunately, there are actual higher [10] which means applications must take locality into account when choosing their target NVDIMMs. This is obviously true for 1LM because NVDIMMs are accessed by the processor like DDR.…”

Section: Locality Of Nvdimmsmentioning

confidence: 99%

See 1 more Smart Citation

Opportunities for Partitioning Non-volatile Memory DIMMs Between Co-scheduled Jobs on HPC Nodes

Goglin

Proaño

2020

Euro-Par 2019: Parallel Processing Workshops

View full text Add to dashboard Cite

The emergence of non-volatile memory DIMMs such as Intel Optane DCPMM blurs the gap between usual volatile memory and persistent storage by enabling byte-accessible persistent memory with reasonable performance. This new hardware supports many possible use cases for high-performance applications, from high performance storage to very-high-capacity volatile memory (terabytes). However the numerous ways to configure the memory subsystem raises the question of how to configure nodes to satisfy applications' needs (memory, storage, fault tolerance, etc.). We focus on the issue of partitioning HPC nodes with NVDIMMs in the context of coscheduling multiple jobs. We show that the basic NVDIMM configuration modes would require node reboots and expensive hardware configuration. Moreover it does not allow the co-scheduling of all kinds of jobs, and it does not always allow locality to be taken into account during resource allocation. Then we show that using 1-Level-Memory and the Device DAX mode by default is a good compromise. It may be easily used and partitioned for storage and memory-bound applications with locality awareness.

show abstract

Section: Locality Of Nvdimmsmentioning

confidence: 99%

“…When NVDIMMs are exposed as additional NUMA nodes, we implemented in hwloc a way to find out the corresponding local CPUs and DDR by looking at NUMA distances and memory target-initiator information in Linux. 10 Figure 10 depicts an example of such configuration.…”

Section: Dax Localitymentioning

confidence: 99%

Opportunities for Partitioning Non-volatile Memory DIMMs Between Co-scheduled Jobs on HPC Nodes

Goglin

Proaño

2020

Euro-Par 2019: Parallel Processing Workshops

View full text Add to dashboard Cite

show abstract

“…It takes a sequence of data page access requests as an input, considering them one by one (1). For each such request, the algorithm retrieves the available page metadata (2), such as the storage tier it is currently located at and whether it is dirty (updated while on a volatile storage tier). Then, the algorithm will consult the data admission policy of the current storage tier on whether the page should be admitted to a different tier (3).…”

Section: A Framework Overviewmentioning

confidence: 99%

“…Existing industrial DBaaS products overwhelmingly offer simple service availability as their SLO 1,2 , with only some offering more advanced SLOs, such as query latency, throughput, etc. 3 Research papers, in addition to the SLOs used in the industry, also consider resource utilization (CPU, RAM, Disk I/O) by tenants [8].…”

Section: E Sla Violation Penalty Functionsmentioning

confidence: 99%

“…A recent development in data storage technologies (dating back to the 2015 Intel and Micron announcement of the 3D XPoint technology [1]) has disrupted the traditional two-tier storage and BP manager architectures and designs. Non-volatile memory (NVM) is a novel family of data storage manufacturing technologies that promise and, to some extent, already delivers [1,2] storage that is almost as fast as RAM, persistent, byteaddressable, and nearly as abundant and as cheap as disk.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Persistent Memory-Aware Buffer Pool Manager Simulator for Multi-Tenant Cloud Databases

Gruenwald

d’Orazio

Leal

2020

2020 IEEE 36th International Conference on Data Engineering Workshops (ICDEW)

View full text Add to dashboard Cite

Non-Volatile Memory (NVM) is a promising development for Database Management Systems (DBMS), offering abundant and fast storage to complement traditional disk and main memory architectures. NVM introduces additional data migration possibilities to the traditional buffer pool (BP) managers used by the DBMS. Hence, the efficient use of this new technology requires a redesign of the BP manager. For the cloud Database-as-a-Service products, this need for a redesign is further complicated by the traditional cloud providers' goal to minimize the Service Level Agreement (SLA) violation penalties paid to their tenants. Unfortunately, current research in the area does not provide a comprehensive picture of the components constituting a multi-tenant persistent memory aware BP manager for a cloud database that makes use of NVM. Furthermore, researchers lack the software tools needed to quickly prototype and estimate the effectiveness of novel data management policies guiding those components. In this paper, we attempt to remedy both issues, first, by proposing a generalized framework that defines the purpose and the abstract interfaces of various multitenant persistent memory-aware BP manager components, and second, by developing and demonstrating a simulator algorithm that is shown to aid in quick testing of different implementations of those BP manager components.

show abstract

Differentiated Services Oriented Auction Mechanism Design for NVM Based Edge Caching

Zhang

Liu

Cai

et al. 2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Initial Experience with 3D XPoint Main Memory

Cited by 9 publications

References 6 publications

Opportunities for Partitioning Non-volatile Memory DIMMs Between Co-scheduled Jobs on HPC Nodes

Opportunities for Partitioning Non-volatile Memory DIMMs Between Co-scheduled Jobs on HPC Nodes

A Persistent Memory-Aware Buffer Pool Manager Simulator for Multi-Tenant Cloud Databases

Differentiated Services Oriented Auction Mechanism Design for NVM Based Edge Caching

Contact Info

Product

Resources

About