Performance: “Screaming”

July 25, 2024

Ivan Ischenko

StarWind Director of Product Management. Ivan is an expert in virtualization and storage architecture. With deep knowledge of software-defined storage and data protection, he provides technical leadership in solution design and product strategy. Ivan delivers high-authority insights into modernizing enterprise-scale IT infrastructure and optimizing virtualized ecosystems.

Tiny overhead, lowest latency, and reduced costs with efficient resource usage and a lightweight 2-node setup.

Intro

Recent advancements in hardware, including CPUs, RAM, NVMe drives, and networking, have unlocked a wealth of potential performance that can be theoretically achieved. However, we still face some limitations, particularly at the storage layer.

Problem

Imagine this scenario: You’ve invested in the latest hardware, complete with all-NVMe storage, cutting-edge CPUs, and the fastest networking. But when you put it to use, you notice some overhead and limitations that leave you dissatisfied. This frustration is understandable—you’ve paid for the hardware and you want to maximize its potential.

Many of the storage protocols used in the industry today aren’t optimized for modern hardware, leading to multiple bottlenecks at various levels. Some are based on the TCP protocol, which can limit performance and increase CPU usage that could otherwise be allocated to your production workload, especially in HCI configurations. Others have internal mechanisms that aren’t efficient enough, effectively reducing performance.

While there have always been limitations in technology, the nature of these limitations has shifted over time. Two decades ago, the performance of hardware often lagged behind the capabilities of legacy storage protocols such as iSCSI, SMB, and NFS, requiring vendors to optimize their implementation to match hardware capabilities. Today, advancements in hardware have outpaced these legacy protocols, making it essential to adopt newer protocols and technologies to fully leverage the high performance of modern hardware.

Solution

StarWind VSAN can assist you in tapping into the full potential of your storage subsystem using the NVMe-oF protocol. We provide a top-tier, highly available (HA) storage architecture that offers enhanced performance on the same hardware, lower latency, and a superior CPU/IOPS/$ ratio.

Leveraging RDMA technology, we eliminate the need for the storage subsystem to constantly use the CPU and Networking TCP/IP stack by directly interacting with the remote server memory.

This type of storage architecture allows you to achieve greater performance with less hardware. StarWind NVMe-oF HA enables you to establish a two-node HA storage system without switches, thereby reducing your capital expenditures (CapEx).

Furthermore, with the NVIDIA-powered GRAID SupremeRAID, you can further reduce the CPU usage on each server (which can be used for your production workload) and boost performance even more.

Conclusion

In terms of storage subsystem performance, NVMe drives are the market’s fastest option. As prices for NVMe drives continue to drop and the understanding that they represent the future of high-performance systems grows, an increasing number of people are investing in hardware that will remain relevant for the next five years. However, many are encountering bottlenecks with standard storage configurations. With StarWind’s implementation of NVMe-oF HA, you can finally unleash your hardware’s full potential and achieve outstanding performance, even on two nodes.