Download the authoritative guide: Enterprise Data Storage 2018: Optimizing Your Storage Infrastructure
Here are two key facts to know before buying enterprise flash storage: First, there is a lot of energy around buying more flash for the corporate data center. Second, as of 2018 only 10% of corporate data centers own an all-flash array.
This matters because it is important not to get ahead of ourselves in the buying decision. Flash in hybrid arrays has a higher install rate, flash is making inroads into the data center, and flash adoption rates are rising. At the same time, this is nowhere near a done deal for all-flash data centers.
There is a huge installed user base of good old fashioned hard drives, and tape for that matter. This is not a simple matter of waiting until the old media lifecycles run out and refreshing with flash. Flash development is proceeding very quickly, but for now at least there are cheaper and better media for storing nearline and long-term data. Right now, nothing beats cheap high-capacity disk or tape for cold storage.
Although flash may work well for active archives (truly active, not just name only) its 3 to 5-year durability window does not make it a trustworthy media for storing long-term data.
Why Companies Buy Flash
A key driver of flash investment – and where you will see development and purchases rising rapidly – is flash for high production performance. This includes mission-critical transactional databases, analytics and high-performance computing workloads, virtual desktop infrastructure (VDI) and streaming assets for the broadcasting and medical industries
So, when we discuss enterprise flash storage we are not talking about the mythical Phoenix of the all-flash data center. We are talking about three things:
1. All-flash arrays for high performance data processing on the shared storage side. These arrays tier from highest performance flash to higher capacity SSDs in the same array. Over time IT should tier the data off the AFA to less expensive media. When they were first introduced a few years ago, FA products tended to be retrofitted hybrid arrays or purpose-built models that lacked the mature data intelligence tools of HDD and hybrid arrays. Over the years AFAs have significantly improved their native data intelligence including compression and D dedupe, native encryption, thin provisioning, and sophisticated management tools.
2. A subset of AFAs are called flash appliances. They are all-flash storage systems, but unlike mainstream AFAs they have only basic storage intelligence. For example, most data center AFAs come with native dedupe, compression, snapshots, replication thin provisioning, and more.
The flash appliance only has basic storage management features in order to consume a minimal amount of processor overhead. They are strictly for boosting performance with high IOPs, low latency and fast throughput. This allows data center admins to gain very fast performance at a lower cost, since the array is engineered for speed and users do not pay for sophisticated management on top of fast flash.
3. Server-side flash for the highest possible performance and latency requirements. Server-side flash is essentially direct attached storage on steroids. It is not the optimal choice for many types of data, but it is for extremely high-performing applications.
Server-based PCIe Cards
Accelerates single application server.
Example: Oracle Financials database.
Accelerates multi-host application.
Example: Docker environment.
High performance storage array.
Example: NAND and SATA flash tiers combine performance and capacity.
Persistent storage or cache for high performance server application.
Single-digit TB capacity lower than AFA or flash appliance.
Data protection requires mirroring datasets, which significantly raises prices.
Combines fast performance and high capacity in the 10s of TBs.
Simple feature set requires 3rd party functionality for data protection and high availability features.
High performance networked storage with intelligent data services.
Efficiently processes mixed workloads.
Dedupe and compression optimizes capacity.
Choosing Among Types of Enterprise Flash
Different Types of Enterprise Flash
With that in mind, let’s talk about common types of flash, AFAs and server-side flash. We'll also look at use case for each.
NAND is nonvolatile storage that does not require power to retain data. The most widely used type of flash in the world, it is installed in both in consumer devices and enterprise SSDs. NAND comes in many different flavors and development is largely driven by increasing density. Single level cell (SLC) was the start and referred to one bit per cell. Greater density followed with multi-level cell and three-level cell architecture.
Today, developments include 3D Vertical NAND, or V-NAND flash chips. By stacking cells in vertical layers, the NAND increases SSD performance and durability with less energy requirements. NAND is the primary flash type for all-flash arrays, although not the only one. NAND AFA primary usage cases are high performance storage systems and tiered AFAs that replace slower flash/HDD hybrids.
SAS and SATA SSDs
SAS and SATA are the same connection protocols we all know and love from our HDD days. SATA SSDs are generally slower than SAS SSDs and are used for high capacity storage systems that need a boost in performance over HDD high capacity systems. Flash SATA can be manufactured in a DIMM form factor and fit into also into a DDR3 DIMM slot as a server boot device and cache.
SAS SSDs are considerably more expensive than SATA SSDs but deliver higher performance and more functionality. SAS SSDs use the SCSI interface and dual port architectures. IT often uses flash SAS to map fast drives to different controllers for multi-path I/O and controller failover. They also have more native management tools and usually come with dual ports, which lets IT map each drive to two separate controllers for the fail-over and multi-path IO that are often required in enterprise storage.
PCI Express (PCIe)
PCIe is server-side flash that can act as a high performance SSD or server cache. When IT connects a high-performance SSD drive to the PCIe interface, data transfer rates can achieve 252Gb/s and IOPs of 1 million-plus.
PCIe lowers resource overhead and the requirement for HBAs and drive controller components. PCIe is specific to the server (i.e. Direct Attached Storage) so trades off DAS disadvantages for high performance features. Popular use cases include mission-critical transactional databases.
Non-Volatile Memory Express (NVMe)
NVMe is a flash spec that enables SSDs to use the PCIe bus to improve performance and lower energy usage. The specification operates in between random-access memory (RAM) and SSDs. The operating system can access it as a storage device, or as memory if the card resides in memory slots.
NVMe drivers combined with the PCIe interface results in very low latency under heavy and mixed read/write workloads. This is an especially helpful architecture for massively scaled databases whose performance improves with server-side flash.
Flash Purchase Decision Points
The major decision points for flash purchases include both technology and business features.
Technology: On the technology side, the foundational flash type will make a difference. Even market-leader NAND presents choices between SLC, MLC, EMLC, TLC and more. Device interconnects like PCIe, NVMe, SAS flash and SATA flash are also important to balance performance, capacity and location in the data path.
Use Case: Ironically, the last concern for some IT buyers should be the first: the use case. VDIs, enterprise data center, streaming web servers, high performance monetized archives: they take different flash decisions.
Costs and ROI: Costs will also influence your choice. Flash prices are dropping but HDDs are significantly less expensive. As storage pros know, HDDs are not as fast as flash, but perform quite well and, for a lower cost, provide more than adaquate value.
Warranties and maintenance: It's important to consider your vendors’ warranties and maintenance pricing. Flash energy costs are lower than HDDs, which offsets some of flash devices’ higher purchase price. But if your vendor offers a 2-year warranty instead of 5-year or gives you a low introductory maintenance price then jacks it up in years 2-5, you’re looking at an expensive proposition.
Before You Buy, Consider These Factors
No matter what your use case for flash, there are 9 critical factors to weigh when you are researching your flash purchase: capacity, performance, availability, durability, scalability, support, management, simplicity and connectivity. Not every flash environment will require top features but know what you are buying and what tradeoffs you are willing to make.
1. Capacity. Understand raw capacity, usable capacity and average data reduction ratios with dedupe and compression. Tell the vendor the types of workloads you are likely to use with flash, since different workloads dedupe with very different ratios.
2. Performance. Performance entails three measurement: IOPs, throughput and latency. As with dedupe ratios, your performance numbers will differ across different workloads. IOPs are the number of block operations per second, latency is how fast the data transfer begins, and throughput is the measure of how fast each block is processed.
3. Availability. 99.999% is the most common enterprise measurement. Asked to see detailed tests and background, and how their flash architecture achieves this level of availability.
4. Durability. Specifically ask what durability features the SSDs come with. It should certainly arrive with garbage collection, packet striping, wear leveling and error code correction. If it does not, then you probably bought a cheap SSD but will have to spend more money on third-party products to protect them.
5. Scalability. No fewer buying scale up or scale out. With the former, CO far you can scale capacity without hitting performance barriers. If the latter, find out how expensive it is to add SSDs for linear performance and capacity scaling.
6. Support. Ask where the support is located, and if engineers can help or just read from a script. Also ask how much support is built into the AFA. Ask about native monitoring services that send alerts about performance impacts, failing SSDs and reaching capacity thresholds. Also go over maintenance contracts with the proverbial fine-comb. You do not want to grab onto a lowball price quote for one to two years, only to watch your operational budget double or triple thanks to support costs.
7. Management. Determine that your AFA comes with native intelligence like replication and snapshots, monitoring and alerts, detailed reports, self-healing, application and platform integration, de-dupe and compression and caching and tiering.
8. Simplicity. Don’t trade heavy IT overhead for performance gains. Watch for automated management functions including simple firmware upgrades, hot replacements, policy-based management and linear scalability.
9. Connectivity. Connectivity should serve you now and in the future. Look for multi-protocol connections and OS support. Consider both iSCSI and FC connections. Check that for high-speed network connections.
For more questions and answers about flash purchases, see 12 Questions to ask All-Flash Vendors.
Next Steps in the Flash Buying Decision
Will the all-flash data center dream come true? It depends on how you define the data center. If the center is production data only, then all-flash is fair game. If you are retaining backup and archives, then all-flash is a poor way to go.
But you don’t have to hold out for an all-flash data center to invest in all-flash arrays and server-side flash. Like any other IT investment, buy for the usage case. And shop carefully!