The customer is a global technology company operating high-load digital platforms across multiple business domains. Its infrastructure supports large-scale, data-intensive services where databases, analytics pipelines, background jobs, and customer-facing applications need to remain responsive under heavy traffic.
In this environment, storage performance directly affects application latency, database query execution time, service availability, and the ability of engineering teams to scale without adding unnecessary operational complexity.
For this kind of infrastructure, the storage layer is not just a backend component. It directly influences how reliably applications perform during traffic peaks, maintenance operations, backup workloads, and other data-intensive processes.
Storage Challenge
The customer’s primary challenge came from the limitations of standard Linux mdraid in high-load database environments.
The company used NVMe-based storage for workloads where consistent throughput and low latency are critical. However, the existing mdraid-based setup created several issues:
- inconsistent drive latency;
- frequent I/O spikes across the system;
- high I/O wait times;
- inability to fully utilize the performance potential of NVMe drives;
- significant write performance degradation on parity-based RAID levels;
- slow rebuild times;
- limited flexibility for scaling and resizing RAID arrays.
The biggest problem was that fault-tolerant RAID configurations came with too much performance overhead. RAID0 could provide speed, but without redundancy. Parity-based RAID levels, such as RAID5 and RAID6, provided protection, but in the customer’s previous setup they introduced a significant write performance penalty.
For high-load database infrastructure, this tradeoff was not acceptable. High-load systems must stay responsive during traffic peaks, maintenance tasks, background jobs, and backup workloads. Any storage instability or lack of performance can quickly surface as higher application latency or reduced database performance.
The customer needed a data protection solution that could combine NVMe performance, predictable latency, fault tolerance, and operational flexibility on bare-metal Linux servers in hosted infrastructure environments.
Solution
The customer deployed Xinnor xiRAID as the RAID engine for its NVMe-based software-implemented data protection.
The goal was to keep the simplicity and control of bare-metal installations while removing the storage bottlenecks the customer had experienced in previous storage deployments. xiRAID was used underneath the existing Linux storage stack, exposing a standard block device and allowing the team to continue using familiar components such as LVM, LUKS, and XFS.
The benchmark environment included:
- CPU: 2 × 2.1GHz Intel Xeon Platinum 8474C, Sapphire Rapids
- Memory: 32 × 128GB Hynix DDR5 4800 ECC REG
- NVMe drives: Micron 7500 PRO MTFDKCC15T3TGP
- Operating system: Debian 12
- Kernel: Linux 6.1
From an operational perspective, the deployment remained straightforward. The customer added license monitoring and alerting on top of the xiRAID installation, with no need to redesign the application layer.
The key requirement was not simply to make RAID0 faster. The customer needed to make fault-tolerant RAID practical for high-load database workloads. With xiRAID, the team was able to use RAID5 and RAID6 while maintaining strong NVMe performance and avoiding the severe write penalties previously seen with mdraid parity configurations.
Test Results
The customer ran a mixed random read/write FIO workload on a filesystem using LVM, LUKS, and XFS.
| Testcase | Read Bandwidth, GB/s | Random Read KIOPS | Write Bandwidth, GB/s | Random Write KIOPS |
|---|---|---|---|---|
| mdraid0 lvm, luks, xfs filesystem |
10.9 | 169.5 | 3.6 | 56.5 |
| XiRAID5 Lvm, luks, xfs filesystem |
12.4 | 194.1 | 4.2 | 64.7 |
In this test, xiRAID RAID5 outperformed the previous mdraid RAID0 baseline while also providing fault tolerance.
Compared with mdraid RAID0, xiRAID RAID5 delivered approximately 15% higher performance overall. This result was especially important for the customer because it demonstrated that the infrastructure team no longer had to choose between performance and reliability. xiRAID allowed the customer to use a parity-based RAID level while exceeding the performance of its previous non-redundant mdraid configuration in the tested workload.
Operational Benefits
After moving to xiRAID, the customer improved the stability and scalability of its storage layer.
The most important benefits included:
-
Smoother disk utilization
xiRAID helped reduce I/O spikes and improve the consistency of disk performance under load. -
Lower I/O wait times
By removing the mdraid bottleneck, the customer was able to use its NVMe hardware more efficiently and improve the behavior of database workloads under pressure. -
Practical fault-tolerant RAID
The customer could use RAID5 without the severe write performance penalty previously experienced with mdraid parity RAID. -
Better scalability and flexibility
The ability to resize RAID arrays on the fly gave the infrastructure team more flexibility as database workloads continued to grow. -
Improved behavior during rebuilds
Faster and more predictable RAID behavior helped reduce the operational impact of rebuild processes on database environments. -
Simple bare-metal operations
xiRAID fit into the existing Linux-based infrastructure and required only license monitoring and alerting on top of the installation.
Conclusion
By replacing Linux mdraid with Xinnor xiRAID, the customer removed a critical storage bottleneck from its high-load database infrastructure.
The result was a better balance between performance, redundancy, and operational flexibility. In the tested mixed random read/write workload, xiRAID RAID5 outperformed the previous mdraid RAID0 baseline by approximately 15% overall, while adding fault tolerance.
For the customer’s infrastructure team, this demonstrated that NVMe storage could be used more effectively in production-oriented database environments without forcing a tradeoff between speed and data protection. xiRAID enabled the customer to unlock more of the performance potential of its NVMe hardware while maintaining the reliability and scalability required for modern high-load digital platforms.
