When discovering new technology, one of the exciting aspects is learning about the future value it offers. When Vexata, Inc announced a new range of storage appliances, the future potential was always in being able to expand the architecture past the initial two controller model. With the announcement of VX-Cloud, Vexata is making good on the promise of the original VX-OS architecture and design.
As we adopt new technologies like NAND flash and persistent memory, the design of shared storage needs to change. Traditional architectures have bottlenecks at both the front and back-end of their designs. Dual controllers simply can’t exploit the performance of flash. So, modern designs need to scale out to use the performance and capacity available. However, scalability is hard, because data coherency has to be maintained. This places constraints either on design or requires a different approach to maintaining state.
- The Race towards End-to-End NVMe in the Data Centre
- In Storage, Software is the New Hardware
- Why Deterministic Storage Performance is Important
In the Vexata architecture, front-end connectivity and back-end storage are disaggregated from each other and connected via a high-speed mid-plane, in this case, lossless Ethernet. This allows front-end connectivity to hosts to be scaled on-demand and storage performance and capacity to be scaled at the back end of the platform. In the initial hardware instantiation of VX-OS, Vexata used front-end I/O control modules (IOMs) and back-end ESMs (Enterprise Storage Modules) to deliver these capabilities. Front-end performance is accelerated using commodity FPGAs, while storage state is maintained across ESMs.
Theoretically, both IOMs and ESMs can be scaled at will, although VX-100 systems currently offer only 2 IOMs and up to sixteen ESMs.
As already stated, the future potential of VX-OS was always in being able to scale the architecture. VX-OS features are essentially software-defined, making it possible to implement the solution on commodity hardware. With the announcement of the VX-Cloud architecture, this is exactly what Vexata is promising – the ability to implement petabyte-scale storage using commodity components.
At this point in time, Vexata is announcing VX-Cloud as a tech preview. As such, this means some of the hardware details are not being disclosed. However, we can look at the design and see how solutions could be built. Figure 2 shows the VX-Cloud architecture in more detail. There are three components; Data Acceleration, Data Distribution and Data Aggregation. Data Acceleration nodes are servers that implement the front-end functionality of the VX-100 IOMs. These are essentially servers with commodity FPGA cards running the VX-OS software.
Data Distribution is implemented using lossless Ethernet layer 2 switches that offer 100GbE performance (such as the Mellanox SN3800). Data Aggregation nodes are standard servers with NVMe storage and VX-OS software. Depending on the drive types used, this could mean up to 1PB per server with designs that can use form factors such as Intel Ruler.
- #84 – Discussing New NVMe Form Factors with Jonathan Hinkle
- Flash Capacities and Failure Domains
- Performance Benchmarks: Reading Between the Lines
Naturally, performance and throughput numbers will depend on the specific hardware used. Vexata has a data sheet available online that shows some example performance numbers based on a range of configurations.
The Architect’s View
At this stage, VX-Cloud is in technical preview and we can expect to see more details later this year. I’m pleased to see that what I initially found interesting about the VX-100 architecture is now starting to be delivered as a cloud-scale storage solution. I think we’re entering an era of “SAN 2.0”, where shared storage is adapting to deliver both to the needs of large-scale enterprise and cloud service providers. This could be an interesting time for the industry as end users accept that shared persistent storage is still important.
If you want to learn more about VX-100, check out this paper I authored that covers how VX-100 works and what each aspect of the architecture delivers.
Disclaimer: Vexata, Inc is or has been a customer of Brookend Ltd.
Copyright (c) 2007-2019 – Post #CD42 – Brookend Ltd, first published on https://www.architecting.it, do not reproduce without permission.