NVMe-oF At Scale for Performance-Intensive Applications – E8 Storage Briefing Note

One of the most significant challenges facing the storage industry today is minimizing application latency without breaking the bank, during the shift to premium-priced non-volatile memory express (NVMe) flash storage systems. The lightning-fast levels of solid-state drive (SSDs) performance and dramatically reduced storage latency facilitated by the NVMe interface can quickly overwhelm compute resources, rendering the storage controller a key performance bottleneck. As a result, performance efficiency and capacity utilization drop dramatically as more SSDs are deployed in a storage array. NVMe-over Fabrics (NVMe-oF) is developing to better share resources and improve utilization, but it must be implemented in a way that also preserves performance and maximizes availability and reliability.

E8 Storage – Local Performance with Shared Efficiency

E8 Storage offers a networked flash storage architecture that was designed specifically to take advantage of NVMe and fast-performing remote direct memory access (RDMA) connectivity. According to E8 Storage, it can facilitate latency levels that are within ten microseconds of those of local SSDs, in an NVMe-oF implementation.

E8 Storage describes its architecture as truly shared everything. It decouples the data and control path, and it pools SSDs into a common resource with which application servers can communicate. Performance is accelerated not only because it removes the storage controller bottleneck, but also because input/output (I/O) operations are parallelized and run concurrently (a singular SSD can support multiple application servers). This is particularly important when it comes to fully exploiting the vastly expanded number of commands and queue depths of NVMe as opposed to SAS SSDs. Flash media can “clear” items from the I/O queue very quickly, while making sure that queue is always full is the responsibility of the storage software. Furthermore, network hopping and cross-talk between servers are minimized.

In addition to accelerating performance, E8 Storage’s architecture stands to improve cost efficiencies because compute and storage resources are better utilized, can scale independently, and because granular thin provisioning is applied. The thin provisioning helps to increase the number of workloads that each SSD can support, because it dynamically allocates capacity among workloads based on the drive’s free capacity, and the minimum capacity that the workload requires.

Another value of E8 Storage’s architecture is that there is no single point of failure, due to the nature of its shared approach as well as its rack scalability. Availability and reliability are further enhanced by the fact that data services operate independently; failure of one data service will not impact others, and an outage or migration of an underlying server will not impact data services. The architecture was designed with network multi-pathing for fast failover, and the nodes may be deployed in high-availability enclosures. It utilizes RAID data protection and does not store volatile data on hosts, to prevent against drive or power source failures, respectively.

StorageSwiss Take

NVMe storage systems come at a price premium, in large part because they require more internal compute and memory to take advantage of the levels of performance of which NVMe technologies are capable. Enterprises must find the right balance of performance acceleration, capacity efficiency, reliability and cost to optimize their investments. If it is implemented correctly, NVMe-oF can help enterprises to find this balance.

For its part, E8 Storage effectively facilitates a high-performance NVMe SAN environment that can boost infrastructure utilization, availability and reliability when compared to local SSD deployments, and at the same time harness more of the raw performance of which SSDs are capable. E8 Storage’s value proposition will come into play largely in use cases such as high-performance analytics, genomics processing, and file storage workload acceleration, which must serve very large data sets with maximum levels of I/O throughput. In particular, large enterprises and cloud and managed service providers that have software-defined cloud environments and that are looking to replace local SSDs should consider E8 Storage.

Sign up for our Newsletter. Get updates on our latest articles and webinars, plus EXCLUSIVE subscriber only content.

Senior Analyst, Krista Macomber produces analyst commentary and contributes to a range of client deliverables including white papers, webinars and videos for Storage Switzerland. She has a decade of experience covering all things storage, data center and cloud infrastructure, including: technology and vendor portfolio developments; customer buying behavior trends; and vendor ecosystems, go-to-market positioning, and business models. Her previous experience includes leading the IT infrastructure practice of analyst firm Technology Business Research, and leading market intelligence initiatives for media company TechTarget.

Tagged with: , , , , , ,
Posted in Briefing Note

Enter your email address to follow this blog and receive notifications of new posts by email.

Join 25,553 other subscribers
Blog Stats
  • 1,906,179 views
%d bloggers like this: