How to choose the Best Flash Array: a Guide to SSD hard disk purchase

When purchasing enterprise flash memory, users need to consider capacity, performance, availability, durability, and other very important factors. Its use cases include mission-critical transaction databases, analytical and high-performance computing workloads, virtual desktop infrastructure (VDI), and streaming media assets in the broadcast and medical industries.

Before purchasing enterprise flash, users need to understand the following two key facts: First, it may take a lot of energy to buy more flash memory for the enterprise data center. Second, as of 2018, only 10% of enterprise data centers have all-flash arrays.

This is important because it is important not to exceed your ability in the purchase decision. Flash memory in hybrid arrays has a higher installation rate. Flash products are entering the data center, and the utilization rate of flash memory is increasing. At the same time, for all-flash data centers, this is far from achieving the goal.

Today, there is a large user base of traditional hard drives and tapes. It is not easy to wait for the original traditional media lifecycle to run out and update flash products. The development of flash memory is progressing very rapidly, but there are at least cheaper and better media for storing near-line and long-term data. Nowadays, nothing is more suitable for cold storage than cheap high-capacity disks or tapes.
Although flash memory may be applicable to active files (real active data, not just names), its 3-to-5-year persistent window does not make it a trusted medium for storing long-term data.

Why do companies buy flash memory

People will see rapid growth in development and procurement, and high-performance flash memory. The key drivers of flash investment include mission-critical transaction databases, analytical and high-performance computing workloads, virtual desktop infrastructure (VDI), and streaming media assets in the broadcasting and medical industries.
So when people talk about enterprise flash memory, they are not talking about the unknowable business of full flash data center, but about three things:
　　
(1) All-flash array for High Performance Data Processing at Shared Storage. These arrays are layered from the highest performance flash memory to higher capacity SSD hard disks in the same array. Over time, IT departments should layer data from all-flash arrays to cheaper media. When first launched a few years ago, all-flash storage products were often modified hybrid arrays or dedicated models, lacking mature data intelligence tools for HDD hard drives and hybrid storage arrays. Over the years, all-flash arrays have significantly improved their native data intelligence, including compression and disaster duplicate data deletion, native encryption, streamlined configuration, and complex management tools.

(2) A subset of the all-flash array is called a flash device. They are all-flash storage systems, but unlike mainstream all-flash arrays, they only have basic storage intelligence. For example, most data center all-flash arrays have the functions of native duplicate data deletion, compression, snapshots, replication of streamlined configurations, and so on.

(3)Server-side flash memory for maximum performance and latency requirements. Server-side flash memory is basically direct additional storage on an all-flash array. And it is not the best choice for many types of data, but it is suitable for extremely high-performance applications.

With this in mind, common types of flash memory, all-flash arrays, and server-side flash memory are discussed below and then look at each use case.

(1) NAND flash memory

NAND flash memory is a non-volatile memory that does not require a power supply to retain data. It is the most widely used type of flash memory in the world and is installed on consumer devices and enterprise-class SSD hard drives. NAND flash memory has many different styles, and its development is mainly driven by the increase of density. Single-stage cell (SLC) is the beginning and each cell is called a bit. The density is higher and followed by multi-level unit and three-level unit structure.

Nowadays, development includes 3D vertical NAND or V-NAND flash memory chips. By stacking units in a vertical layer, NAND flash memory can improve the performance and durability of SSD hard disks, while reducing energy consumption requirements. NAND is the primary flash type of an all-flash array, although it is not the only flash type. NAND full-flash array is mainly used in high-performance storage system and hierarchical full-flash array, which can replace slow flash/HDD hybrid storage.

(2)SAS and SATASSD

SAS and SATA are the connection protocols of the hard disk era that people know and like. SATASSD hard drives are usually slower than SASSSD hard disks and are used in high-capacity storage systems that require higher performance than HDD hard disk high-capacity systems. Flash SATA can be manufactured in memory shape, and can also be used as a server startup device and buffer to insert DDR 3D IMM slot.

SASSSD hard drives are much more expensive than SATASSD hard drives, but can provide higher performance and more functionality. The SASSSD hard disk uses a SCSI interface and a dual-port architecture. IT often uses flash SAS to map fast drives to different controllers to achieve multipath I/O and controller failover. They also have more native management tools and usually have dual ports, which enables IT to map each driver to two separate controllers for failover and multipath IO that are often required in enterprise storage.

（3）PCI Express（PCIe）

PCIe is a server-side flash memory that can be used as a high-performance SSD hard disk or server cache. When IT connects a high-performance SSD hard drive to the PCIe interface, the data transfer rate can reach 252 GB / s, and the IOP can reach more than 1 million.

PCIe reduces resource overhead and the demands for HBA and components of drives and controllers. PCIe is server-specific (that is, directly connected to storage), so there is a lack of DAS disadvantages for high-performance functionality. The common use cases include mission-critical transaction databases.

(4) Non-volatile Memory (NVMe)

NVMe is a flash memory specification that enables SSD hard drives to use the PCIe bus to improve performance and reduce energy consumption. The specification runs between random access memory (RAM) and SSD hard disk. The operating system can access it as a storage device, and if the card is in a memory slot, it can access it as memory.

The combination of NVMe driver and PCIe interface can achieve very low latency under heavy and mixed read/write workloads. This is a particularly useful architecture for large-scale extended databases, whose performance improves with server-side flash memory.

The decision point for flash memory purchase

The main decision points for purchasing flash memory include the function of technology and business.　　

Technology: In technology, the basic flash type will make a difference. Even NAND flash memory, the market leader, also provides the choice between SLC, MLC, EMLC, TLC and other products. The interconnection of PCIe, NVMe, SAS flash and SATA flash is also important for balancing the performance, capacity and location of data paths.

Use Case: Ironically, the last concern of some IT buyers should be the first use case. VDI, Enterprise Data Center, Streaming Media Web Server, High Performance Monetized Archives: They all take different flash decisions.
　　
Cost and return on investment: Cost can also affect the choice of the enterprise. The price of flash memory is falling, but the price of hard drives is significantly lower. As storage professionals know, HDD hard drives are not as fast as flash memory, but they perform fairly well and provide more than enough value at a lower cost.

Warranty and maintenance: it is important to consider the warranty for vendor and pricing maintenance. The cost of flash energy is lower than that of hard drives, which offsets higher purchase prices for some flash devices. However, if the supplier of the enterprise offers a 2-year warranty instead of a 5-year warranty, or provides you with a lower entry-level maintenance price, you will see an expensive proposition to upgrade it within 2 to 5 years.

Before purchasing, please consider these factors

Regardless of the use case for flash memory, there are nine key factors to be weighed in the study of purchasing flash memory: capacity, performance, availability, durability, scalability, support, management, simplicity and connectivity. Not every flash environment requires top-level functionality, but the enterprise knows what to buy and what trade-offs it is willing to make.

(1) Capacity. Understanding raw capacity through duplicate data deletion and compression, it is available to reduce ratio using capacity and average data. Tell vendors the type of workload that may use flash memory because different workloads reuse different ratios.

(2) Performance. Performance requires three measurements: IOP, throughput, and latency. As with the deletion rate of duplicate data, performance numbers vary from workload to workload. IOP is the number of block operands per second, delay is the speed at which data transmission begins, and throughput is a measure of the processing speed of each block.

(3) Availability. 99.999% is the most common enterprise measurement standard. It requires a detailed look at the test and background, and how their flash architecture achieves this level of availability.

(4) Durability. Ask about the durability of SSD hard drives. It's bound to come with garbage collection, data stripping, loss balancing, and error code correction. If not, you might buy a cheap solid-state hard drive, but you have to spend more on third-party products to protect them.

(5) Scalability. Increase in the scale of purchases and expanding the scale. For the former, flash memory can be extended far more without performance barriers. If the latter, please understand the cost of adding SSD for linear performance and capacity expansion.

(6) Support. Ask about the support location and whether the engineer can provide help or just read from the script. And it is necessary to ask how much support is built into AFA. Ask native monitoring services that send alerts about performance impacts, SSD hard disk failures, and reaching capacity thresholds. And the maintenance contract should also be completed by combing. Companies do not want a low offer for a year or two, just to support twice or three times their operating budget, which is due to support costs.

(7) Management. Determine that the enterprise's all-flash array has native intelligence, such as replication and snapshots, monitoring and alerts, detailed reporting, self-healing, application and platform integration, duplicate data deletion and compression, and caching and layering.

(8) Simplicity. Don't spend a lot of IT overhead on performance improvements. Focus on automated management features, including simple firmware upgrades, hot replacement, policy-based management, and linear scalability.

(9) Connectivity. Connectivity should serve the present and future of the enterprise. Find multi-protocol connections and OS support. Consider iSCSI and FC connections and check for high-speed network connections.

Next steps in Enterprise purchase decision

However, enterprises do not have to support all-flash data centers to invest in all-flash arrays and server-side flash memory. As with any other IT investment, use cases should be carefully selected when purchasing.

Will the dream of an all-flash data center come true? It depends on how the enterprise defines the data center. If the center is just producing data, all flash memory is a fair game. But all flash memory is a bad way to keep backups and archives.