Tech Leadership

Scaling In-Memory Data Processing with Samsung Advanced DRAM and NAND/SSD Solutions

on Aug 04, 2017

Samsung and VoltDB analyze advances in memory and in-memory database technologies to address challenges from massive increases.

In-memory computing is here to stay. After a decade of development, it has come of age at a most opportune time. In-memory has evolved from solutions that offered low throughput and low latency, to systems that offer high throughput and impressively lower latency, and now to systems with exceptionally high-throughput reads and even lower latencies.

Three fundamental factors have been driving this trend:
• Hardware is now available that can use higher amounts of DRAM (High Capacity DRAM) – up to 3 TB – as compared to the previously existing barrier of around 512 GB RAM per server
• SSDs have become a ubiquitous solution for all datacenter storage needs. While SATA SSDs cannot meet performance SLOs (service level objectives) at the higher end of the application spectrum, this can be accomplished with advanced NVMe storage technology and Samsung’s recently developed Z-SSD (5X faster than NVMe).
• The benefits of very large servers with high-density DRAM and high-speed SSDs can be fully attained when systems are designed to take advantage of them.

VoltDB: An In-Memory, Real-Time Decision-Support System

VoltDB has advanced the role of an OLTP database into that of a real-time DSS (Decision Support System). It has already enabled this transformation in a variety of use cases. These include cases involving:
• real-time portfolio risk monitoring and management
• credit card fraud detection and prevention
• monitoring of sensor data for anomaly detection
• monitoring of network management data
• several elements of the Evolved Packet Core (mobility management – MME, user authorization and authentication – AAA, policy management – PCRF/PCEF, Least Cost Route calculations, real-time billing – OCS/OFCS)
• other key functions of telecommunications companies
• hyper-personalized customer interaction applications typically found in the online gaming and advertising industries.

There is a new class of applications and use cases that requires ultra-fast interaction with the database, within the context of a larger dataset than would be traditionally considered “memory friendly.” These applications require single-digit millisecond latency across data sets in the 10s of terabytes.

Today’s conventional scale up vs. scale out data processing is now evolving into a postmodern definition of scale up AND scale out. Change agents pushing this transformation range from Single Source of Truth (combining multiple ODSs) initiatives in traditional financial service enterprises, as well as the increasing desire of ecommerce organizations to require extremely fast data collection, and to significantly improve next-generation filtration systems for the Industrial Internet of Things (IIoT).

Samsung has been spearheading a rapid evolution in chip architectures (including those pertaining to High Capacity DDR4, NVMe SSD and NAND/Z-SSD) that is enabling a critical advancement in the design of in-memory data processing systems like VoltDB. These advanced memory developments are able to scale to handle ever more complex applications with larger datasets and greater volumes of contextual data, without compromising data safety or recovery time SLOs).

High Capacity DRAM

Samsung High Capacity memory solutions are well exemplified by today’s most advanced DDR4 technology. Modules comprised of Samsung DDR4 are designed with new system circuit architecture to deliver significantly higher performance with lower power requirements than previous generation of memory products.

Doubled bandwidth, along with reduced voltage and more efficient power consumption, not only improves performance, but also optimizes total cost of ownership. Furthermore, the deployment of Samsung 8Gb DDR4, produced using an advanced 10nm-class process, assures an optimal level of performance, greater bandwidth, sizeable power savings and higher reliability in the manufacturing of high capacity DIMMs.

Today, Samsung DDR4 DRAM offerings range from 8GB capacity points to 128GB capacity points. Samsung DDR4 LRDIMM (load reduced memory module) technology uses a distributed buffer to maximize memory bandwidth efficiency when scaling to higher capacities and speeds. In general, RDIMMs are more power-efficient, while Load Reduced DIMMs show an improvement in speed.

As a result, the combination of LRDIMM’s features and Samsung 8Gb DDR4 chips produced using a 10nm process node provides an optimized solution for highly virtualized environments, high-performance computing and networking.

PM1725 NVMe SSD

To satisfy the exceptionally high demands of enterprise environments, SSDs must perform over long periods at their maximum levels to accommodate the variety of workloads simultaneously accessing the devices. In addition, the selected SSD must provide consistent performance that will satisfy highly demanding Quality of Service (QoS) requirements. The Samsung PM1725 SSD, which was selected for these tests, is designed to excel in virtually any datacenter scenario. This enterprise-level, ultra-high-performance SSD provides unsurpassed random read performance and is particularly well suited for read-intensive datacenter applications. When compared with the other standardized SSDs in Samsung internal testing, the PM1725 delivers the highest random read IOPS.

Conclusion

Today’s rapid evolution of data utilization patterns is leading to greater amounts of data being used for faster processing, and even transactional work.

Samsung has been leading efforts to enable the creation of larger-footprint servers that will decrease datacenter operational expenses, with its high-density DRAM and ultra-fast NVMe SSD storage drives.

For more information including VoltDB and Samsung Benchmark results, please refer to the PDF file below.

LIST

Related insights