xiRAID is the world’s first and fastest software RAID for the new generation of storage, that not only protects your directly attached flash drives, but also those connected via NVMe-oF™. Learn more why xiRAID performance is superior to any hardware RAID controller, and the level of reliability is even greater.



Korzystanie z xiRAID zamiast tradycyjnego sprzętowego kontrolera RAID zapewnia naszym klientom dodatkową wydajność pamięci masowej, lepszy całkowity koszt posiadania i prostszą konserwację systemu.

  1. Much faster than hardware RAID controllers
  2. Independent from hardware upgrades and modifications
  3. Simple migration between storage systems
  4. Saves one PCIe slot (for even more NVMe drives or extra connectivity)
  5. Unlimited product lifetime
  6. Immune to physical damage
  7. Works with NVMe-oF™ devices


High Performance

  • 150 GB/s and 30 000 000 IO/s
  • Latency below 0.5ms
  • No performance penalty in mixed I/O workloads
  • Low performance hit in degraded RAID state

Usability and compatibility

  • Wide range of RAID levels:
  • 0/1/10/5/6/7.3/50/60/70
  • Works with local and remote (NVMe-oF™) drives
  • Clients access RAID array as a local block device
  • POSIX API support: no need to change applications or file systems


  • Less than 10% CPU usage under maximum workload
  • Low system memory footprint: less than 4 GB of DRAM required for optimal performance


xiRAID is offered in two versions: xiRAID Classic, designed for kernel space, and xiRAID Opus (Optimized Performance in User Space).

xiRAID Classic

  • Operating within Linux kernel, requiring compatibility updates with kernel updates.
  • Suitable for local RAID.
  • Exports a linux block device.
  • Supports all RAID levels, rebuild, and more.
  • Is managed through CLI.
  • Utilizes a small portion of all available CPU cores distributing load evenly.
  • Works on x86 CPU only.

xiRAID Opus

  • Operates in user space, independently of the kernel, simplifying kernel updates and maintenance.
  • Suitable for network devices or virtualization, without creating local RAID.
  • Can be operated via virtIO, NVMeoRDMA, NVMeoTCP.
  • Offers additional built-in features like NVMe initiator, NVMe over TCP/RDMA, iSCSI target, and Vhost controller.
  • Provides a distributed CLI for managing multiple servers.
  • Requires few specific dedicated CPU cores and fully occupies them.
  • Support both x86 and ARM architectures (DPU).



NVMe can no longer be considered a new technology – it’s been here for almost a decade and the past few years saw an ever-increasing adoption rate for NVMe storage devices both in the consumer and enterprise markets. Legacy SAS and especially SATA protocols are becoming bottlenecks for SSDs, and a transition to a faster, more efficient storage technology has very compelling TCO and performance benefits.

Composable Disaggregated Infrastructure (CDI) approach is also rapidly emerging, bringing the abstraction of compute, storage, and networking resources from their physical locations. All of these lead us to a new world where we need new approaches for performance and data protection.

However, moving from such a well-known technology and ecosystem like SATA and SAS that have been with us since 1970’s, is not easy. Some market research states “…performance problems associated with storage software inhibit the growth of the market. Furthermore, issues related to data integrity and data protection act as challenges for the growth of the market.”


Xinnor is positioning the innovative xiRAID as the only software solution on the market today that effectively works with local and networked NVMe devices. With I/O handling parallelization and lockless datapath, xiRAID arrays have very small RAID penalties and perform very close to the raw hardware capabilities. xiRAID is purpose-built for new types of flash drives, with a strong attention to their architecture and datapath features. It is a lightweight software module with low CPU and RAM usage.

xiRAID works with block devices – local or remote, using any transport – PCIe, NVMe-oF or SPDK target, by Fibre channel or InfiniBand. And it provides a local block device to the system.



Compared to the rich ecosystem of SATA and SAS RAID options, working with NVMe is not so straightforward. First, there are hardware RAID adapters that can RAID NVMe devices as well as SAS/SATA. These cards are fast enough for 4 drives directly attached, but for more they have to rely on PCIe switches with added performance hit due to PCIe lane oversubscription. They add hardware cost, another layer of complexity between the CPU and drives, and have performance limitations due to SoC capabilities (1.8M IOPS per card).

Software RAID offers high flexibility, zero associated hardware costs and vendor-agnosticism in terms of compatibility. It’s worth noticing that software RAID is currently the only option to support the new class of NVMe-oF JBOF (EBOF) devices for disaggregated storage in the CDI world.


  1. We have our own patented RAID calculation engine, based on a decade of math research and in-depth knowledge of CPU operation.
  2. Lockless datapath, no spinlocks, no scheduling, no memory copy.

For fastest RAID we designed a new calculation engine, which is fast and simple. It uses only one simple vector operation and has less data move operations.

Need a demo or quote?

Feel free to contact us at info@abart.pro

Tell us which version of XINNOR xiRAID software you need and for how many drives in the array.