A web-scale data center needs to scale to support thousands if not millions of users. Web-scale application use cases range from traditional transactional IT to modern data analytics and machine learning workloads. It also includes new advanced video applications and traditional HPC workloads. While the workload type may vary, each of these workloads has universal demands of the storage infrastructure: low latency, and high performance.
Enterprise data centers are left at a crossroads when trying to identify the right storage system as the foundational component of that storage infrastructure. In most cases, these storage systems have a parallel file system placed on top of them that is appropriate to the workload.
The problem is the traditional storage systems enterprise IT is familiar with don’t perform to the extreme level these applications need. An alternative approach is for the storage system vendor to attempt to reduce latency and increase performance by integrating proprietary hardware like custom ASICs or FPGAs. However, integrating custom hardware components limits choice and flexibility.
The storage system in these web-scale environments needs to do more than be flexible. It also needs to work with the data no matter where the location. In the web-scale data center, some workloads are massively parallel, and some are not. The storage system needs to deliver a consistent combination of throughput, IOPS, and latency.
Excelero – High Performance without Hardware Lock-in
Excelero targets its NVMesh solution directly at the web-scale use cases. NVMesh is a software-defined, low overhead, ultra-low latency, block storage system on which customers typically place a parallel file system. It provides local flash performance and low latency across an NVMe network. The customer, who more than likely has a significant investment in server hardware to run applications, can continue to use that hardware with the Excelero software.
Excelero has two deployment options. Both deployment options provide a single unified pool of storage, rotating parity and linear scaling. The first deployment option is a converged configuration. The NVMesh target and client software run on all nodes. The other deployment option, and thus far the most popular, is top-of-rack flash. In this configuration, the NVMesh target runs on dedicated storage nodes. Then the NVMesh client runs on the customer’s application servers. Because of Excelero’s low latency and the NVMe Fabric, the applications see similar performance to that of local storage, without the downsides.
Since Excelero is block storage for web-scale data centers, users or application owners are free to lay down the file system of their choice on top of it. In the latest release NVMesh v2, Excelero provides features that enable the customer to use their solution for a variety of workload types.
Excelero is continually making refinements large and small to optimize performance, but the NVMesh v2 release is more focused on broadening the appeal of the solution. NVMesh 2 adds a capability called MeshConnect. Excelero’s NVMesh now supports almost any fabric, including NVMe over Ethernet, NVMe over Fibre Channel and Infiniband. It also supports a variety of protocols including TCP/IP, RDMA and FC.
The expansion in network protocol support gives Excelero two advantages. First, customers looking to try out the solution no longer have to implement a new network type and protocol (NVMe over Ethernet). Instead, they can implement the solution on their existing network infrastructure. Utilizing the existing infrastructure allows them to test the solution before investing in a new network infrastructure.
The second advantage is it gives Excelero the ability to land and expand. The organization can purchase the solution to solve a specific performance challenge and then amortize that investment on other applications that could also benefit from a performance boost.
The second prominent feature in NVMesh2 is MeshProtect. In version 1, NVMesh protected data by either mirroring data or striping and mirroring data. While ideal for performance-sensitive applications, it did consume capacity. In NVMesh 2, Excelero introduces optional erasure-coding based parity protection. In an 8+2 (eight drives data and two parity drives) MeshProtect delivers 80% usable capacity while maintaining data access even if two drives fail. Excelero focused on providing MeshProtect without sacrificing performance. It leverages Intel’s AVX support built into modern CPUs to help speed up parity calculations.
The third feature of NVMesh 2 is MeshInspect which provides more in-depth insight into capacity and performance utilization. The new analytics dashboard provides insight into cluster-wide performance and utilization. Customers can customize the statistics that appear on their dashboards.
Excelero is the basic building block of a storage architecture designed to support the modern data center. Traditional applications can directly access the block storage volumes, and modern applications can access it through the parallel file system of their choice. It is particularly ideal for analytics and high-resolution post-production workloads. HPC environments can use Excelero as a burst buffer. Also, modern container environments can count on Excelero for consistent scalable, high-performance persistent storage. With version 2.0 of NVMesh, Excelero broadens the use cases to more traditional applications by supporting more common protocols and delivering a more capacity conscious media protection model.
Sign up for our Newsletter. Get updates on our latest articles and webinars, plus EXCLUSIVE subscriber only content.