ESG Lab Validates Virident Benchmarks

Virident Systems, Inc., in flash-based Storage Class Memory solutions, announced that Enterprise Strategy Group Lab validated benchmarks created by Virident to measure enterprise flash-storage performance.

The ESG Lab brief documents the challenges of benchmarking enterprise SSDs using current tools and concludes that Virident’s General Public License (GPL) licensed SSDBench benchmark utility is suited to address the challenges associated with predicting the performance of SSDs with real-world application workloads.

"If your organization is using or evaluating flash-based solid-state storage, you should download SSDBench and take it for a test drive," said Brian Garrett, VP at ESG Lab. "As you consider the benefits of this valuable and relatively expensive new technology, you owe it to yourself and your organization to perform a thorough analysis of your investment."

In the brief, ESG Lab documents the challenges of accurately benchmarking SSDs that are driving the need for the new benchmark methodologies included in SSDBench.

The challenges include:

• Speed: The biggest challenge associated with benchmarking solid-state storage is the difference in speed compared to traditional disk drives.
• Conditioning: Flash-based solid-state performance changes over time. To account for this, SSDs need to be conditioned with reads and writes to achieve a steady state. Performance testing of a flash device without conditioning yields results that are higher than users will see in production over time.
• Wear Leveling: A flash memory cell supports a finite number of writes. Flash controllers address this issue with a set of spare erased cells and algorithms that spread writes as evenly as possible over all the cells in a device. This technology, which is known in the industry as wear leveling, impacts write performance over time. In the worst case, a poor wear-leveling algorithm can lead to a dramatic drop in performance for write-intensive workloads that is referred to in the industry as the ‘write cliff.’
• Variability: Performing I/O at the speed of flash introduces a number of performance and tuning challenges for storage architects. Performance of SSDs can vary over time due to a number of factors, including the queue depth being exercised and the size of I/O requests.
• Real-world Workloads: Storage professionals recognize the need to test performance with a combination of low-level characterization tests and real-world application workloads. Characterization testing with only reads or only writes is a poor indicator of performance for real-world applications, which tend to be a mix of reads and writes.

"Virident created SSDBench because we found our customers didn’t have an industry-accepted means to effectively and accurately measure SSD performance," said Shridar Subramanian, VP of product marketing at Virident. "We believe there should be a common benchmarking methodology that models realistic application behavior for SSDs so that customers can appropriately weigh the pros and cons of the multitude of solutions there are in the market. SSDBench is an effective way for customers to determine, prior to their purchase, if an SSD solution they are considering provides adequate performance to ensure a strong ROI."

About SSDBench
SSDBench is a freely available GPL benchmarking tool that Virident built for solid-state application workload performance analysis. It leverages open source benchmarks including FIO, IOZone and INNOSim. Characterization tests run for hours instead of minutes and employ a variety of block sizes and queue depths, with a goal of characterizing performance over time. Application workloads evaluate SSD performance for a variety of real-world applications that benefit from the speed of flash (e.g., analytics, high-frequency trading and OLTP databases). A freely available spreadsheet template automates the graphical presentation of results.

More on ESG SSDBench Brief (registration needed)