Why Hardware RAID Cannot Cope with NVMe Performance: Interview with Sergey Platonov

February 7, 2024

Back to all posts

In our new blog post, we had the opportunity to speak with Sergey Platonov, our VP of Product Strategy, who brings a deep understanding of RAID technologies to the table. Throughout the interview, Sergey discusses the significant challenges and limitations inherent in hardware RAID solutions, from the constraints of PCI-Express lanes to logistical issues highlighted by the COVID-19 pandemic. This conversation offers a thoughtful exploration into the future of RAID, guided by Sergey's expertise in the field.


Interviewer: Hi Sergey! Today, we're discussing hardware RAID solutions and their issues. Let's start with the basics. Could you please describe the main problems with hardware RAID?

Sergey: Certainly. The issues with hardware RAID can be divided into two main categories. The first is related to the limitations of the PCI-Express bus. A hardware RAID is connected through the PCI-Express, typically having 16 lanes. This means connecting more than 4 NVMe drives becomes inefficient due to oversubscription, as each NVMe uses 4 lanes. This leads to a bottleneck in the bus's performance.

The second category of issues is related to physical and economic constraints. Implementing more powerful controllers becomes economically impractical due to the limited space on the RAID adapter and the need for heat dissipation. There are also difficulties in accommodating additional memory and a supercapacitor for power loss protection.


Interviewer: So, can hardware RAID scale efficiently?

Sergey: Actually, hardware RAID is limited to the maximum performance of four drives. Want to connect more? You'll need to add additional RAID controllers, complicating the architecture and reducing overall system efficiency.


Interviewer: Are the manufacturers motivated to develop hardware RAID further?

Sergey: I don’t think so. The limited opportunities for performance improvement and physical constraints make innovation in this area challenging. Manufacturers might improve speed and rebuild times to maintain current standards, but significant changes are not expected.


Interviewer: Sergey, NVMe drives are known for their high performance, directly connecting through PCI-Express. Can you elaborate on why hardware RAID configurations might be less optimal for NVMe compared to traditional SAS/SATA drives?

Sergey: Absolutely. NVMe drives leverage the full bandwidth of PCI-Express lanes to deliver unparalleled performance. This presents a challenge for hardware RAID configurations, which are often bottlenecked by the number of available PCI-Express lanes. While SAS/SATA drives, being inherently slower due to interface limitations, can be efficiently managed by hardware RAID controllers without saturating the bus, NVMe drives quickly exceed these limits, leading to underutilization of their potential. This is where software RAID solutions become crucial, offering the flexibility to harness the speed of NVMe drives more effectively without being constrained by the physical limitations of hardware RAID controllers.

SAS vs. SATA vs. NVMe drives comparison


Interviewer: Considering the significant performance differences, how does the efficiency of NVMe drives in a RAID setup compare to that of multiple SAS/SATA drives? Is there a scenario where traditional drives could be more advantageous?

Sergey: Theoretically, even a few NVMe drives in a RAID configuration can outperform a larger array of SAS/SATA drives due to their superior speed and lower latency. However, SAS/SATA drives may still hold an advantage in scenarios requiring high capacity storage with lower cost per gigabyte, or where the sheer speed of NVMe is not critical. Furthermore, traditional drives can offer more mature and stable hardware RAID solutions in environments where NVMe's cutting-edge performance is not a necessity or where budget constraints are a consideration. It's all about aligning the storage solution with the specific needs and goals of the IT infrastructure.


Interviewer: Are there any more issues of hardware RAID?

Sergey: There’s also such a significant issue as problems with supply chain. Especially during the pandemic, the production and supply of many components faced serious delays. Some vendors had delivery times that could reach 7-8 months. This posed substantial difficulties for companies, especially if they urgently needed to expand or replace equipment. The sudden demand for online services affected the situation as everyone rushed to procure equipment in large volumes.

One more crucial thing is having no support for network-connected drives. Hardware RAID controllers are generally designed to work with locally connected disks. This means they do not support disks connected via a network, limiting their use in disaggregated storage configurations. Modern data centers and cloud providers often use disaggregated architectures to increase flexibility and resource management efficiency. The lack of support for such configurations significantly reduces the attractiveness of hardware RAID solutions for these scenarios.


Interviewer: It seems these limitations significantly affect the choice between hardware and software RAID solutions in modern IT infrastructures.

Sergey: Absolutely. While hardware RAID has its advantages, such as dependability and broad operating system support, software solutions offer greater flexibility and better meet the requirements of modern data centers, especially regarding performance, scalability and support for disaggregated storage systems.