A mathematics and IT management specialist from the University of Calgary, John Schouten embarked on his career in the IT industry through work to support applications and hardware based upon early x86 systems. Over the past fifteen years, he has focused primarily on infrastructure solutions related to server consolidation, data centre optimization and the availability of data centre assets, and now serves as enterprise technology strategist at Dell. In his current position, John has responsibility for the positioning and development of go-to market strategy for the enterprise portfolio, and acts as focal point for large data centre projects.
In this guest contribution presented below, John offers a seasoned perspective on new approaches to storage management that balances use of revolutionary technologies with enterprise need for various storage media and cost control. Through tiering and hybrid arrays, he concludes, enterprises can "get the flash performance they need, when they need it," while "staying closer to the cost of a disk solution."
The exponential growth of data puts enormous demands on storage infrastructures. According to IDC, businesses’ storage demands are growing in excess of 50 percent a year while their total available storage capacity is growing at barely half that rate, presumably due to cost constraints.
Meanwhile, virtualization and the explosion of I/O-intensive applications for Big Data analytics, database transactions, and more have inspired in IT leaders and their end users a desire for higher levels of storage performance. CIOs are having to choose between performance and cost, and increasingly, those choices are becoming more challenging to call.
Consumed by a ‘new is better’ mentality, the IT industry has a tendency to discard what seems to be ‘old’ technology, even when it continues to provide useful service. Unfortunately, a complete ‘rip and replace’ strategy can be more expensive and a lot more complex. It makes the most sense to add new technology, like flash and tiering, to an existing infrastructure to address new demands, improve performance or lower costs for existing applications.
Technologies such as flash, solid-state memory and tiering have evolved to the point where, when combined, provide the speed, agility, and cost-efficiencies needed to offer a good alternative to a full rip and replace approach. And when it is deployed intelligently, flash performance is available to businesses at the price of disk today.
Flash is fast
Flash is the new high performance leader, which comes in a various forms, each delivering significant advantages over high end hard disks (HDDs). Disk is not going to disappear anytime soon, but it will be relegated to less demanding work over time.
Flash benefits range from cost for performance — the dollar per IOPS (input/output operations per second) is more than seven times cheaper for flash than for HDDs — to lowering rack space and power consumption costs. Beyond the infrastructure, flash also boosts employee productivity and helps meet business critical SLAs.
When IOPS is the key metric, then flash clearly outperforms disk. While a conventional 15k HDD can deliver approximately 200 IOPS, a single SSD can provide thousands of IOPS in the same form factor.
The two most popular flash technologies are single-level cell (SLC) and multi-level cell (MLC, or enterprise-class eMLC). SLC is ten times more persistent, with three times faster sequential write, comparable sequential read, and more than four times the cost of MLC. Flash storage also comes in various formats and is being deployed in both all-flash and hybrid — a mix of flash and HDD — models, as well as inside servers (i.e. PCIe cards).
According to Forrester, enterprise adoption is growing with 30 percent already using solid-state storage and another 32 percent planning to deploy it. Forrester expects that flash will become ubiquitous in transaction-heavy environments, not just performance-sensitive ones, in the near future.
Disk is not dead
Though all-flash storage excels in high-performance use cases, disk and hybrid systems will continue to serve major roles in the data centre. Disk will be around for at least another decade and will continue to complement flash-based storage for applications such as databases and email and in the support of virtual server environments.
While all-flash arrays compare favorably to enterprise arrays with high performance hard drives, they do not compare favorably to high capacity hard drives.
Unstructured data growth dictates the need for dense, bulk storage of less critical data that disk drives most affordably support. Unless thousands of people are accessing the same file at the same time, as in a web fronting application, hard drives still make sense.
Tiering delivers the best of both worlds
Tiering enables CIOs to seamlessly bridge the price/performance chasm and assign data and applications to the most appropriate storage medium. It involves assigning different categories of data to different types of storage media to ensure optimal performance and the lowest total cost.
Tiering can be considered the equivalent of an automated workflow that knows which ‘packages’ of data require ‘immediate express delivery’ and places them on that first tier (in this instance flash) and which ‘packages’ of data can be safely stored on less expensive and slightly slower second tiers. Not only can tiering allocate data between the different media, but innovative vendors have developed the capability to also automatically allocate data across write-intensive SLC and read-intensive MLC SSDs. This improves performance for data-intensive applications and workloads in high-performance storage solutions that can achieve over 300,000 IOPS.
A storage array’s ability to automatically tier across multiple SSD drive types is revolutionary and offers numerous advantages. While many available flash arrays leverage write-intensive SLC drives, a balance of MLC and SLC offers customers greater overall cost for performance. Overall flash reliability is increased when an array leverages the more vulnerable MLC flash tier mostly for reads. In this approach, capacity in the more expensive SLC tier can be kept to a minimum, at levels just high enough to handle inbound write traffic. As a result, this model dramatically reduces the overall cost to implement flash.
The attraction of all-flash arrays is the predictable nature of performance. With all-flash, businesses do not have to worry about a tier or cache miss causing data to be served from hard disk, but without tiering, this comes at a high cost.
To be clear, all flash (with or without tiering) does not solve all storage challenges on its own. The latest all-flash array offerings on the market today typically lack full enterprise-class features (e.g. advanced replication, replays and management) and the industry integrations that the more established vendors provide. There are also significant cost savings to be gained through adding new capabilities to a business’ existing storage environment, and avoiding the costs associated with the ‘rip and replace’ strategy noted above.
A storage infrastructure that allows easy morph into a hybrid array -- one that mixes SLC, MLC and disk -- can further reduce costs and increase capacity, offering a much lower price point per/GB than all-flash arrays, while providing the performance of flash. As a result, businesses get flash performance when it’s needed, and do so at a price that’s comparable to an all-disk solution.
While flash adoption is growing, the real value lies in tiering that optimizes every application and every volume to best address both price and performance metrics. Tiering offers businesses the best of both worlds: data is written to the fastest tier using SLC drives and as data ages, the data is automatically moved to MLC drives, and eventually to slower and much less expensive traditional HDD drives.
By embracing this innovative approach and adding new capabilities to existing storage infrastructure, businesses are finding they can get the flash performance they need when they need it, and achieve improved storage performance while staying closer to the cost of a disk solution.