Are the Performance Gains of Optane SSDs Cost Justified?

By Hubert Yoshida posted 10-02-2020 03:11

Hitachi’s Unified Compute Platform (UCP) HC is a Hyperconverged (HCI) appliance solution that combines VMware’s vSAN software and vSphere and vCenter, running on clustered, x86 servers.

UCP HC maximizes performance via Intel Optane SSDs. Optane replaces traditional NAND flash technology with 3D XPoint technology, which combines the density of NAND flash SSDs with the speed and addressability of DRAM memory. Whereas NAND must erase and write a 2KB-16KB block before writing, each 3D XPoint location can be individually addressed and written. This results in an SSD providing consistent read and write latency regardless of write activity, and performance comparable to system memory. UCP HC uses NVMe to ensure fast and efficient movement of data to traditional and Optane SSDs. NVMe (non-volatile memory express) is a communications standard/protocol developed specially for non-volatile memory to use the PCIe bus, enabling SSDs to act more like the fast memory that they are, rather than the hard disks they imitate. NVMe supports massive parallelization—up to 64k commands within a single I/O queue, and 64k I/O queues per device. These features enable vast amounts of parallel I/O operations over the node’s PCIe bus, reducing bottlenecks and keeping data flowing to UCP HC’s multiple processor cores. However, there is a cost associated with this versus standard SATA SSD.

Last year Hitachi Vantara did a test on our hyperconverged UCP HC system focused on
performance for virtualized environments comparing SATA and NVMe SSD Configurations. Testing was designed to demonstrate performance and cost-efficiency using industry-standard tools and methodologies. This test was audited by The Enterprise Storage Group, ESG, and the results were published in an ESG Technical Review of March 2019. Hitachi Unified Compute Platform HC: Maximize Hyperconverged Performance with NVMs.   

Testing compared two four-node Hitachi Unified Compute Platform HC (UCP HC) clusters. The first cluster used UCP HC V121F SATA-SSD nodes while the second cluster used UCP HC V124N all-NVMe nodes. Storage was provided by VMware vSAN software running on each cluster.

ESG evaluated UCP HC in a typical enterprise OLTP environment using and Industry standard, Open source database load testing and benchmarking tool, HammerDB. This workload emulates the data base activity of a typical online brokerage firm as they generate trades, perform account inquiries, and execute market research. The workload is composed of ten transaction types with a defined ratio of execution. Four of the transactions performed database updates, and the rest were read only. HammerDB measured the number of new orders per minute (NOPM) which were entered into the database.

Although the NVMe configuration was 16% more expensive than the SATA configuration, at maximum performance with four VMs, the NVMe configuration achieved 46% better performance and demonstrated a 26% advantage in the number of transactions per dollar. Increasing the number of VMs to eight, demonstrated an even higher advantage with an 89% advantage in performance and 63% advantage in the number of transactions per dollar.

Increases in the performance of IT infrastructure enables increases in revenue, productivity, and customer and user satisfaction. Improved performance is the most important factor driving the deployment of solid state storage. Organizations considering the deployment of Hyperconverged Infrastructure would benefit by investigating how Hitachi UCP HC with NVMe can cost effectively meet their workload performance requirements.