Virident was founded in 2006 with a simple goal of maximizing CPU utilization by boosting I/O performance. Virident's flash journey began with NOR products, but an eventual restructuring brought them into the NAND space. Their first PCIe SSD, the TachIO, rolled out in 2010 sporting ultra-durable SLC. MLC easily addresses endurance for 95% of applications, and acceptable performance and better economics led to the arrival of the MLC FlashMAX design in 2011. Virident delivers on density and performance, but their true strength lies in wide interoperability with leading applications and an ecosystem of software that enhances application performance. Virident's software offerings leverage the server, storage, and network resources of converged datacenter infrastructures to deliver SAN-like capabilities.
Virident came under the umbrella of storage industry heavyweight WD when HGST brought Virident in October of 2013. HGST is fast becoming the flash arm of WD, and the existing Virident product line fills a gap in their offerings. Virident gains access to an established channel and more engineering muscle to develop tightly integrated server-side flash storage and software.
The HGST FlashMAX II is a Storage Class Memory (SCM) product designed to maximize density and performance, and minimize latency, to provide higher performance in cloud computing, virtualization, database, analytics, HPC, and webscale applications.
The third generation FlashMAX II design features formidable density packed into the HHHL (Half-Height, Half-Length) form factor. Three models, Standard, Performance, and Capacity, address varying performance and density requirements. The Standard model features the lowest speed and densities of 550 GB and 1.1 TB. The Capacity model has incrementally slower speed than the performance variant, but offers a whopping 4.8 TB of storage in the slim HHHL form factor.
The 2.2 TB Performance model, which we are evaluating today, supplies large block random-read bandwidth of 2.7 GB/s, and sequential write speeds of 1 GB/s. The real strength lies in the tuning for transactional workloads, with 1,130,000 512B and 340,000 4k random read IOPS. Sustained 4k write IOPS are spec'd at 110,000. Mixed performance is also impressive with 220,000 75%/25% random read/write 4k IOPS on tap. HGST prides themselves on offering predictable and consistent performance during the warranty period when the drive is full and in steady state. Endurance metrics are impressive at 7.5 DWPD (Drive Writes Per Day) for five years, or 15PB per TB of capacity.
The Storage Class Memory (SCM) distinction is the pillar of the FlashMAX series architecture. The inefficiencies of the SCSI and AHCI driver stack are well known and there are new emerging standards, such as NVMe and SCSI Express, that are developed from the ground up to maximize the driver stack for non-volatile flash. HGST's SCM architecture minimized the driver stack even before finalization of standards such as NVMe.
HGST's Virident Flash-management with Adaptive Scheduling (vFAS) is Virident's own proprietary software layer that streamlines flash access. vFAS virtualizes the underlying NAND components and presents them as a standard block storage device. vFAS sidesteps other storage protocols and interconnect layers and addresses flash more as an adjunct to memory. This results in memory-like 20-microsecond latency.
Many competing PCIe SSD designs employ multiple ASICs assigned to control separate 'drives' (consisting of onboard NAND components) that are controlled by a central RAID controller. This creates a flash hierarchy that limits NAND control to the ASIC managing each bank of NAND. vFAS utilizes an FPGA to provide a global view of the underlying flash resource for the host-based SMC software. Applications can create hot spots of activity that lead to premature wear, but the vFAS architecture maximizes endurance with a global wear distribution technique.
Data integrity is always a concern and vFAS addresses this with a flash-aware RAID scheme and end-to-end data protection. This allows isolation of the onboard NAND components into separate stripes in a RAID 5 (7+1) configuration. The global view of the NAND yields the benefit of a flash-optimized parity scheme that protects from multiple component failures, and capacitors protect from host power-loss events.
Most PCIe SSD implementations, with the exception of Fusion-IO (HGST's chief competitor), offload all SSD functions onto the device itself. This alleviates any host overhead and minimizes CPU cycles dedicated to managing the flash resource. HGST's approach utilizes host processing of NAND management to achieve lower latency than competing devices. This removes other performance bottlenecks and gives the SCM architecture a tight integration with their other software offerings.
HGST does not lead with chart-topping performance numbers, instead focusing on providing a performance and low latency that most users can utilize in a typical environment. Today we test the HGST FlashMAX II against two tough competitors, the MLC Micron P420m, and the SLC Micron P320h. First, let's take a look at the software offering of the FlashMAX II.