We use cookies to personalize your site experience.
Privacy policyFinance
In the high-stakes world of finance, every microsecond counts. High-frequency trading (HFT) firms demand database platforms that deliver rock-solid consistency, ultra-low latency, and linearly scalable throughput under blistering transaction rates. At Xinnor, we architect storage and database accelerators optimized for HFT workloads—so you can focus on your trading strategies, not your infrastructure.
xiRAID Advantages for Finance
Low Read/Write Latency
Cutting-edge NVMe-optimized RAID engine cuts I/O delays to the bare minimum, ensuring your order books stay current to the nanosecond.
Linear Throughput Scaling
Seamless scaling across dozens of virtual machines and physical servers lets your cluster grow as your trading volume surges—no “hot-spot” bottlenecks, no performance cliffs.
Resilient Degraded-Mode Performance
Even when drives fail or network blips occur, our solutions maintain high throughput, eliminating the risk of trading disruptions during critical market windows.
Resource-Efficient Virtualization
Leverage virtual machines for flexibility and cost-control without sacrificing “bare-metal” I/O performance—optimize CPU and storage utilization across your entire compute estate.
Simplified Management & Consolidation
One scalable software RAID engine across NVMe pools dramatically reduces complexity, streamlines provisioning, and slashes total cost of ownership.
xiRAID: The Ideal Storage Engine for HFT & Databases
Xinnor’s xiRAID Opus is a software RAID solution purpose-built for NVMe storage in demanding, virtualized environments. Key features include:
- Purpose-Built for NVMe. Unlike generic Linux mdadm, xiRAID Opus bypasses the kernel I/O stack, harnessing SPDK’s polling model for predictable I/O cycles, even under heavy parallel workloads.
- Efficient CPU Utilization. Leverages polling-mode drivers to reduce CPU interrupts and maximize host core utilization—even under heavy I/O loads.
- Flexible vHOST Integration. Exposes RAID arrays as vHOST targets for seamless VM attachment, enabling consistent performance across multiple database instances.
- Accelerated Rebuild & Degraded-Mode Logic. Offloads parity rebuild operations to background threads and prioritizes foreground I/O to maintain high throughput during drive failures.
- Scalable Pool Management. Combine hundreds of NVMe devices into shareable storage pools, dynamically resize RAID sets, and simplify provisioning with a unified command‑line interface.
Case Study: PostgreSQL Database in Virtualized Environment
Challenge:
PostgreSQL’s transactional robustness is ideal for analytics and order processing—but only if paired with storage that keeps up. Virtualized environments add flexibility but introduce latency overhead. Bare-metal deployments underutilize resources. Xinnor was tasked with finding the sweet spot: virtualized PostgreSQL with “bare-metal” speed.
Solution:
- Benchmarked two RAID approaches on 10x KIOXIA KCMYXVUG3T20 NVMe drives across 14 Ubuntu 22.04 VMs (dual AMD EPYC 9534, PostgreSQL 15, pgbench, 100 clients).
- Compared Linux mdadm vHOST against SPDK vhost-blk backed by xiRAID Opus.
- Measured both normal and degraded (one-disk-down) performance across TPC-B-like and read-only workloads, monitoring with iostat.
Results:
| Test | 1 VM Performance | Latency the lower, the better | Total Server Performance xiRAID: 14 VMs, mdraid: 16 VMs |
|---|---|---|---|
| select-only, xiraid | 110K tps | 0.9 ms | 1540K tps |
| select-only, mdraid | 76K tps | 1.3 ms | 1216K tps |
| select-only, xiraid, degraded mode | 112K tps | 0.9 ms | 1568K tps |
| select-only, mdraid, degraded mode | 4.6K tps | 22 ms | 74K tps |
| simple-update, xiraid | 26K tps | 3.8 ms | 364K tps |
| simple-update, mdraid | 4.3K tps | 23 ms | 69K tps |
| tpc-b-like, xiraid | 21K tps | 5 ms | 294K tps |
| tpc-b-like, mdraid | 4K tps | 25 ms | 64K tps |
- Read-Only (Degraded Mode): xiRAID Opus achieved up to 25x better throughput than mdadm, ensuring near-peak performance even during failures.
- Small-Block Writes & CPU-Intensive Loads: xiRAID delivered 6x faster small writes and 5x higher TPC-B-like throughput compared to mdadm.
- Under single-disk failures, xiRAID maintained ≥80% of peak performance, whereas mdadm throughput dropped sharply—posing risks for time-sensitive trading.
- Higher VM density and predictable I/O enabled consolidation of bare-metal silos into one virtualized pool, translating to up to 40% savings in hardware and overhead.
By partnering with KIOXIA and leveraging Xinnor’s xiRAID Opus, PostgreSQL deployments on VMs achieved near-bare-metal performance, rock-solid degraded-mode resilience, and streamlined resource utilization—empowering HFT firms to transact faster, safer, and more cost-effectively than ever before.
Read the full whitepaperContact Us
Unlock your trading edge with Xinnor’s HFT-optimized database solutions. Contact us to schedule a performance deep-dive and see xiRAID in action in your finance environment.