Democratizing High Performance Data
The data revolution is transforming our world at a stunning pace – and we’re just getting started. It’s now fueling a wave of compute-intensive applications that harness large data volumes to detect patterns, learn, interpret and respond in real-time. These cognitive applications are leading us into a new era of computing, requiring mainstream adoption of exponentially faster compute and storage infrastructure.
The good news is that over the past decade we’ve seen more than 100X increase in compute performance for ML/AI, real-time analytics, and large-scale simulations. These performance advances are a result of ongoing architectural and development ecosystem innovations in CPUs, GPUs, and FPGAs. Indications are that we’ll see another 100X in compute performance over the next decade.
Unfortunately, the progress in data storage I/O performance has been relatively modest. While all-flash arrays brought us an initial 10X performance jump versus HDD storage arrays, we’ve seen mostly incremental performance improvements since then. Even the recently introduced NVMe flash arrays from today’s leading vendors are only marginally faster than their previous SAS/SATA SSD versions.
To compensate for this mismatch in server and storage performance, application developers implement caching grids or expensive hot storage tiers. Aside from the complexity and significant overhead of such approaches, they don’t address the core application requirement of data performance and scale.
What’s needed is a unified, resilient data storage infrastructure that can affordably deliver ‘tier 0’ performance at big data scale. Fortunately, today’s industry-standard SSD technology provides the building blocks to achieve this outcome. What’s been missing is the system and software architecture that can take full advantage of SSD performance and economics. Until now.
At Vexata, we invented an enterprise storage system that achieves the next 100X in performance gains at mainstream all-flash storage economics. Our goal was to democratize high performance data for the cognitive era.
Great, But How?
Today’s leading enterprise all-flash array architectures are a significant advance over spinning disk storage arrays. Yet, their controller architectures still limit them to ‘lighting up’ less than 10% of the SSDs in the array at a time. Plus, they’re not architected for low-latency, random R/W across large data sets – a critical requirement for cognitive era applications.
To realize the next 10-100X in performance across large data sets from these same SSDs requires a completely new architectural approach. So, we began with a clean sheet of paper, an amazing team of engineers, and a clear understanding of the requirements of enterprise storage.
What we invented was a completely new solid-state storage system and software architecture that combines three primary concepts:
- FPGA-based datapath acceleration
- Extreme system-level I/O parallelism
- Local SSD I/O scheduling algorithms
Vexata VX-OS Architecture
This architectural approach has provided Vexata strategic advantage in a number of enterprise environments, simply by providing a cost-effective platform that delivers low latency I/O performance and abundant bandwidth to support the most demanding applications. By using commercially available best-in-breed hardware building blocks, Vexata can deliver this tremendous competitive advantage at cost points that are at or below existing all-flash arrays. This allows Vexata to innovate rapidly, keep increasing performance and density, and support the latest protocols and solution stacks.
The most recent Vexata VX-OS 3.5 is clear evidence that this strategy is paying off. With Vexata VX-OS 3.5, customers will now be able to deploy double the capacity to over 435 TB of usable capacity in a single 6U. By the end of calendar 2018, customers will also be able deploy as much of a petabyte of usable capacity in the same 6u chassis without deduplication or compression.
Vexata VX-100F Scalable NVMe Flash Array
The VX-OS 3.5 release also enables the ability for Vexata customers to select between standard Fibre Channel connections or can begin to non-disruptively migrate to the emerging NVMe over Fabric (NVMe-oF) deployments using either Gigabit Ethernet or Fibre Channel. To learn more about how Vexata is protecting customer investments, attend this webcast titled “A New Era in Data Storage for Machine Learning and AI Ecosystems.”