On Wednesday, August 7, Radian Memory Systems, Inc. will make history at the Flash Memory Summit by demonstrating a live, system-level ‘Apples to Apples’ benchmark comparison for the first time ever between a Zoned Flash NVMe SSD versus a nearly identical, production released FTL-based NVMe SSD.


Targeting compliance with the forthcoming Zoned Namespaces (ZNS) specification from the NVM Express™ industry standards organization, Radian’s RMS-350 Zoned Flash SSD is based upon its Symphonic™ CFM technology that provides idealized, configurable Flash zones that can be associated with namespaces.

The Zoned Namespaces (ZNS) versus FTL-based SSD benchmarking at Flash Memory Summit (FMS) will utilize an overwriting workload and be the most equivalent ‘Apples to Apples’ comparison test ever performed between a Zoned SSD, or any Software-Defined SSD, and a FTL-based SSD.

The fio tester will perform the most widely utilized SSD test workload, involving 70/30 4K Random Read/4K Random Write patterns, contrasting overprovisioning levels, namespaces and other variables while capturing IOPS and latency metrics.

Radian’s Zoned Flash RMS-350 U.2 NVMe SSD will be compared against a production released, FTL-based U.2 NVMe SSD from a leading Tier I vendor based upon identical silicon.  This includes using the same 3D TLC NAND, the exact same Flash array configuration (dies/package, packages/channel, channels and total raw capacity), same DDR array, and the same Flash controller processor.

SDF Benchmarking
Benchmarking Software-Defined Flash (SDF) SSDs, such as Zoned Flash devices, using typical workloads has long been a challenge as these workloads involve overwriting, in-place updates that are not directly supported by SDF SSDs.  Insightful benchmarking must also measure garbage collection, but SDF SSDs might leverage target host systems to control garbage collection.

By emulating a log structured host, Radian’s Block Translation Layer provides Logical-to-Physical address translation, cleaning segments (zones) and relocating valid data.  This not only addresses the SDF benchmarking requirements for garbage collection and serializing in-place overwrites, but provides a direct analog to storage engines found in most purpose-built All-Flash Arrays, Key Value Stores, hyperconverged and Software-Defined Storage frameworks that dominate modern on-premises and hyperscale data centers.

So called ‘Hero’ benchmarks, involving 100% 4K random writes or mean latencies at single queue depth, can make for exciting marketing claims but are of limited use to system designers.  Mixed workloads are of far greater value, and more challenging for SSDs, than 100% read or write patterns.  Similarly, in today’s data centers mean latencies provide an interesting data point, but what matters most is tail latencies (99.99%).  However, even tail latencies can be manipulated through single queue depths, overprovisioning bandwidth, or capacity.

The ultimate performance tests for today’s data center applications show tail latencies (99.99% or greater QoS) and include IOPS in the same measurement, ensuring that bandwidth is not being overprovisioned and that latencies are not being analyzed in isolation.

Another key configuration for today’s data centers involves noisey neighbors.  As virtual machines and multiple client/applications access the same storage devices or arrays, delivering deterministic latency for one application while an adjacent client is undergoing heavy churn is a common requirement.  In public and private clouds this QoS level is often defined as part of a Service Level Agreement (SLA), and has presented a difficult challenge for SSDs.  In addition to single client tests, the Radian ‘Apples to Apples’ benchmark performed at FMS will cover testing with multiple different clients that each run concurrently on different namespaces on the same SSD, quantifying the interference from noisey neighbor traffic.

Live Demonstration
To see the results of this extensive ‘Apples to Apples’ testing, come to the live demonstration on August 7, at the Flash Memory Summit held at the Santa Clara Convention Center at Radian’s Booth #615.