Torwell LLC
Principal-led engineering for performance-critical systems — from bare-metal Linux tuning and storage architecture to DevOps automation and AI workload deployment.
We don't hand you a runbook — we write the automation that makes the runbook unnecessary.
Linux kernel and storage tuning, high-availability architecture, network stack optimization, and hardware-level performance analysis. Built on benchmarks, not assumptions.
End-to-end automation of deployments, configuration management, and fleet operations. We build the tooling that enforces consistency at scale — across hundreds of hosts across multiple datacenters.
Containerized deployment and operation of AI/ML models on NVIDIA GPU infrastructure. Covers AMD64 and ARM server platforms, inference pipeline optimization, and cost-effective on-premise GPU hosting.
Custom monitoring tooling, deep-dive incident forensics, and business-continuity design. We find root causes, not symptoms — and build systems that stay up when things go wrong.
Every engagement starts with measurement. These are representative results from production systems.
A global fixed-income trading platform needed deterministic sub-15 µs round-trip latency on 10GbE. Kernel-bypass networking (OpenOnload), PCI bus re-allocation, and NIC interrupt tuning dropped latency from 29.36 µs to 14.78 µs — without touching application code.
MySQL replication lag on a high-frequency trading firm's primary database was traced to a write-through RAID controller starving disk throughput. Replacing the controller and enabling write-back cache brought IOPS from ~300 to 8,700 — outperforming existing "high performance" nodes by 50%.
A trading infrastructure team had accumulated years of configuration drift across 500+ Linux and Solaris hosts in three global datacenters. We built a validation suite and parallel execution tooling that reduced drift to near-zero — uncovering and fixing 60+ broken installations in the process.
Torwell LLC is a principal-led consultancy founded in 2002 by Alexander Shusharin. We work on systems where performance actually matters — trading platforms, inference servers, high-availability databases, and large-scale Linux fleets.
Alexander spent eight years building and operating the Linux infrastructure behind world-leading fixed-income electronic trading platforms — systems processing hundreds of billions of dollars daily at microsecond latencies. That experience informs every engagement: we measure before we tune, and we automate before we document.
Today Torwell's work spans Linux performance engineering, DevOps automation, containerized AI/GPU deployment, and full-stack web infrastructure — across financial services, media, and technology clients.
Describe your system and the problem you're trying to solve. Initial consultations are free — we'll let you know quickly whether we can help and how.
Typical response within one business day.