Follow Datanami:
November 4, 2013

Top Five Use Cases Where All-Flash Arrays are Most Cost Efficient

Solid state or flash based storage is a more expensive storage media when compared to HDDs on a dollar per gigabyte basis ($/GB). Yet, when intelligently integrated into a system with storage optimization technologies, storage vendors lower the acquisition cost and overall total cost of ownership (TCO). Moreover, $/IO is lowest with the use of solid state storage and dramatically lowers cost in several use cases.  

Remember, flash has 2 inherent benefits, higher performance and greater efficiency:

Higher performance:

  • Multiple gigabytes per second of random data throughput
  • High input/output operations per second:  ~10,000 IOPs (HDD: ~200 IOPs)
  • Consistent response time due to predictable access time

Greater efficiency:

  • Cost efficiency with lowest $ per IOPs
  • Place frequently accessed data, or hot data, on high performing all-flash arrays and less frequently accessed data, or cooler data, on the most cost effective HDDs.
  • Reduces overall storage footprint by avoiding overprovisioning

Let’s look at the first use case:

 1.     Use Flash Instead of Overprovisioning:

Many of today’s techniques to improve storage performance include over provisioning a large quantity of HDDs to achieve high IOPs, which can lead to inefficiencies. In these situations, leveraging all-flash arrays increases IOPS density and reduces spindles, footprint and cooling up to 95% compared the same IOPS performance

with traditional HDD storage media. In the example, a 2U all-flash array can replace up to 2 racks of overprovisioned HDD’s and deliver the same performance.

 NetApp Image

Use flash instead of overprovisioning for new installations and start replacing overprovisioned drives with all-flash arrays.

 2.     Accelerate Applications by Improving Responsiveness

Processor and memory performance have grown with Moore’s Law. When coupled with the increased usage of multi-core processors, virtualization, and accelerated network speeds, but storage performance has lagged behind. Improving the I/O data path and addressing performance bottlenecks is critical to provide the necessary performance throughput. Improving responsiveness of applications is a critical factor. Whether accessing email, an enterprise resource planning (ERP) system, an OLTP database, or performing real time analytics, I/O delays and system latencies cost time and money. Delivering the right data, at the right time, in the shortest amount of time ensures that applications and expensive CPUs are not waiting for data. All-flash arrays deliver extreme I/O performance and very low latency to improve responsiveness of business and research applications.

 3.     I/O Offload in Shared Storage Environments

Shared storage platforms provide tremendous value, however certain workloads can negatively impact performance by consuming valuable compute resources. In many cases, these are temporary workflows that don’t need the efficiency and protection features being offered by the shared storage. All-flash arrays are well-suited to offload the I/O-intensive workloads that most affect latency and user experience. The result is a combined solution which delivers improved performance and responsiveness along with headroom on the shared storage for additional growth.

 4.     Scratch Directories in Lustre

The scratch directories in the Lustre parallel file system are often intended for production work, where extreme performance is needed. The files, which are stored in the scratch space are typically subject to being purged after a number of days. This space is not long-term storage and is accessed with extreme performance by the compute nodes. Depending on the “horse power” or petaflops of the compute nodes, Input Output requirements of 1,000,000 IOPs are not untypical and can very easily be reached with all-flash arrays.

 5.     Hybrid Storage in Data Warehousing

In data warehousing, a typical access pattern is that 80% of the queries are accessing only 20% of the total data in the warehouse. Putting those 20% of frequently accessed data on all-flash arrays solves a lot of issues, such as eliminating I/O bottlenecks, speeding up reports by flattening out peak usage patterns, as well as reducing footprint and energy in the data center.

While there are many positive effects all-flash arrays bring to the data warehouse, probably the most surprising effect is that the overall storage system becomes much more reliable. By putting the “hot” or frequently accessed data on the all-flash array, and the “cold” or less frequently accessed data on HDDs, the HDDs are being accessed less frequently and the overall reliability of the storage system increases.  Because flash memory has no moving parts, failure rates as it’s inherent in the HDD’s, does not exist. Hence this positive side effect has surprised many early adopters of all-flash arrays.

Datanami