The cloud industry’s rapid growth is good for business, but puts cloud providers under great pressure to scale to meet growing and volatile performance demands, while maintaining a viable business model.
Flash memory solves these problems for several cloud environments, including Web-scale and software-as-a-service providers. Growth projections in both Web-scale and software-as-a-service sectors are extremely high. For example, eMarketer analyst Jeffrey Grau projects a $100-billion rise in ecommerce between 2010 and 2015. Gartner (News - Alert) forecasts the total software revenue for SaaS delivery within the enterprise application software market to grow to around 23 billion dollars by 2015.
Web-Scale Up, Not Out
Many Web-scale companies seek to reduce or minimize the Tier 1 storage system costs and meet highly random data access patterns by adopting distributed, scale-out server architectures. The problem with these architectures is that operating costs, including hardware, software licenses, floor space, power and cooling, and IT maintenance and administration soon begin to erode margins, making competitive pricing a challenge.
Additionally, as customer bases and data volume grow, even the most efficient scale-out systems reach performance limits – forcing companies to consider risky and costly architecture overhauls.
NAND flash-based technologies can solve both of these problems. First, it’s possible to integrate flash into servers as an extension to memory so that each server delivers far more performance compared to hard disks – particularly when it comes to the random access patterns common to many OLTP Web environments. Second, it enables Web-scale to reduce aggregate DRAM needs to greatly lower power and cooling costs, particularly in caching tiers. Third, it can completely eliminate the need for performance disk arrays, reducing the number of system failure points and IT overhead.
One social networking company implemented NAND flash in its middle-tier caching farm to cut server footprint by 62 percent (280U of rack space), while eliminating 2,300 disk nodes to monitor and maintain.
Most SaaS providers virtualize and share system resources to minimize costs to keep pricing competitive and protect their bottom line. The challenge many SaaS providers face is that each new customer adds unpredictable workload patterns that quickly reach the limit of disk-based capabilities. This problem is even worse for service providers like healthcare and government that must host multi-tenant databases to ensure all data is independent.
NAND flash-based technologies are ideal for resolving these challenges. They greatly increase per-server workload to decrease infrastructure needs. They are also much more resilient to traffic spikes ensuring SaaS providers meet SLAs without massive build-out.
Additionally, they overcome the primary bottleneck of virtualization – the I/O blender effect, the abstraction layer in virtual environments that makes even sequential I/O random.
There are many examples of SaaS providers using flash technology to improve performance while reducing infrastructure costs. Here are a few cases:
- Host Europe GmbH implemented flash to increase per-server virtualized workloads 2-5x, while eliminating worry of interruptions and slowdowns from traffic spikes.
- Call center workforce management provider, Pipkins (News - Alert), moved its database to flash to exceed 6x per-server workload times, while realizing a payback period of “weeks to months.”
- IBM (News - Alert) Tivoli Server Manager (TSM) SaaS provider, FrontSafe, increased server workloads 3x while cutting its server footprint by 66 percent.
All signs indicate that cloud computing will become a dominant industry in the information technology landscape. But it can’t scale to meet the world’s insatiable thirst for faster processing of growing data volume without more efficient technologies than today’s disk-based systems can provide.
Flash memory provides the performance and capacities that cloud providers need, while at the same time keeping capital and operating costs in check.
Edited by Braden Becker