Turnkey · Spec to rack
Hardware & Infrastructure
Purpose-built ingest servers. Specced for sustained high-throughput data movement, procured from enterprise distributors, assembled and tested in-house, shipped pre-configured. Plug in power and Ethernet — it's live.
We build the machines, not just the software
Most infrastructure software assumes you already have the hardware. That assumption breaks down when the workload is this specific — sustained, multi-lane data ingest from removable storage devices, buffered to NVMe, uploaded over dedicated cloud links, with cryptographic verification at every stage.
General-purpose servers are not designed for this. Consumer USB hubs thermal-throttle under continuous load. Desktop NVMe drives wear out from constant write-delete cycles. Single-NIC configurations have no failover. Off-the-shelf builds leave reliability gaps that surface in production, not in testing.
So we spec, procure, assemble, test, and rack the hardware ourselves. Every component is selected for this workload. Every server goes through a 72-hour burn-in before it ships. When it arrives at your datacenter, the OS is installed, DataBridge Core is deployed, network interfaces are configured, and SSH keys are provisioned. Connect two cables and it joins the fleet.
Reference configuration
This is our standard ingest server spec. Adjusted per engagement based on data volumes, device types, and throughput requirements.
| Component | Specification |
|---|---|
| CPU | AMD EPYC 7313P or Intel Xeon W-2400 series |
| Memory | 64 GB DDR5 ECC (4x 16 GB) |
| Buffer storage | 2x 2 TB NVMe SSD in RAID 0 |
| USB infrastructure | 5-bay USB 3.2 Gen 2 hub (10 Gbps per port) |
| Network | Dual 10GbE NIC (Intel X710 or Mellanox ConnectX-4) |
| Management | IPMI / BMC for lights-out remote access |
| Operating system | Ubuntu 22.04 LTS (headless) |
| Form factor | 2U rackmount chassis |
| Power | Redundant PSU, UPS-backed circuit |
USB infrastructure
The intake point for all physical media. Multi-bay USB 3.2 hubs rated for continuous, sustained throughput across five concurrent device connections.
5 concurrent device connections per server
Each server supports five storage devices simultaneously. SD cards, HDDs, SSDs, CFexpress readers, NVMe in USB enclosures — any block device that presents as removable storage.
Hot-plug capable
Devices can be inserted and removed while other lanes are active. Core detects new devices within seconds and begins the ingest workflow automatically.
Sustained throughput rated
Consumer hubs thermal-throttle under load. We source enterprise-rated USB 3.2 hubs designed for continuous high-bandwidth operation without degradation.
Kernel-level monitoring
USB flap detection built into the kernel log scraper. If a device disconnects and reconnects rapidly, Core pauses the lane and alerts the operator instead of creating a partial copy.
Buffer storage
Two NVMe drives in RAID 0 provide 4 TB of high-speed local buffer. Data lands here before it goes anywhere else.
High-speed local copy first
Data is always copied from the source device to the NVMe buffer before cloud upload begins. This decouples device availability from upload duration — the operator can remove the device as soon as the local copy completes.
Parallel operations
While one lane is uploading from buffer to cloud, other lanes can be copying from device to buffer. The NVMe RAID 0 stripe provides enough bandwidth for all five lanes to operate concurrently without contention.
Integrity manifest generation
SHA-256 hashes are computed during the buffer copy. The manifest is generated before upload begins and verified again after upload via ETag comparison. Two independent integrity checks for every file.
Protection against device removal
If an operator pulls a device during copy, only the buffer is affected — the source device remains untouched. Core detects the interruption, marks the batch as incomplete, and alerts through Watch.
Network
Dual 10GbE or 25GbE NICs connected to datacenter switch fabric and the dedicated cloud interconnect.
Dual NIC with failover
Two independent 10GbE (or 25GbE) interfaces connected to separate switch ports. If one link drops, traffic fails over automatically. No manual intervention, no lost bytes.
Connected to Direct Line
Network interfaces route to the datacenter switch fabric and through to the dedicated cloud interconnect. Data travels over AWS Direct Connect, GCP Interconnect, or Azure ExpressRoute — never the public internet.
Bonding available
For deployments that need more than 10 Gbps per server, NIC bonding aggregates both interfaces into a single logical link. 20 Gbps effective throughput per server.
Reliability
Every component choice is a reliability decision. ECC memory, enterprise drives, redundant power, out-of-band management.
ECC memory
Every bit matters when you are generating cryptographic hashes for data integrity verification. ECC memory corrects single-bit errors before they can corrupt a manifest or file in transit.
Enterprise-grade drives
NVMe drives rated for datacenter write endurance. The buffer sees constant write-then-delete cycles — consumer drives would degrade within months. Enterprise drives are rated for years of this workload.
UPS-backed power
Servers are connected to uninterruptible power supplies. A power event does not mean a lost ingest. Core handles graceful shutdown and resumes on power restoration.
IPMI / BMC remote management
Every server has an out-of-band management interface. Power cycle, BIOS access, console redirect — all without being physically present. Critical for datacenters where our team is not on-site.
Procurement and assembly
Components are sourced from enterprise distributors — not consumer retail. CPUs, memory, drives, NICs, chassis, and power supplies are selected for the specific workload profile, ordered in batch, and tracked through our inventory system.
Assembly happens in-house. Every server is built to the same cable management and labeling standard. Internal cabling is routed for airflow. Front-panel labels identify the machine, its serial, and its network configuration. This matters when you have 10 identical servers in a rack and need to identify one at 2 AM.
After assembly, every server goes through a 72-hour burn-in test: sustained CPU stress testing, memory pattern verification, full-disk write-read-verify cycles on every NVMe drive, and continuous network throughput testing across both NICs. If any component shows degradation or errors, it gets replaced before the server ships.
Ships ready to deploy
Every server ships with the operating system installed, DataBridge Core deployed and configured, network interfaces set up for the target datacenter, and SSH keys provisioned for remote access. The IPMI interface is configured for out-of-band management.
At the datacenter, the installation process is: rack the server, connect power (two cables for redundant PSU), connect Ethernet (two cables for dual NIC). Power on. The server boots, Core starts, and it registers with Watch automatically. No on-site software configuration. No USB drives with install media. No manual network setup.
We have deployed fleets of 10 servers this way. The on-site team needs basic racking skills and a cable guide we provide. Total time from unbox to operational: under an hour per server.
Scalability
Need more throughput? Add more servers. The architecture is horizontally scalable by design. Each server operates independently with its own 5 ingest lanes, its own buffer storage, and its own network uplink. There is no shared state between servers, no cluster coordination, no single point of failure.
One server gives you 5 concurrent device lanes. Five servers give you 25. Ten servers give you 50. The configuration is identical across the fleet — same OS image, same Core version, same network configuration template. Watch sees them all from a single dashboard.
Scaling down is just as simple. Servers can be decommissioned independently. Data is in the cloud — the local buffer is ephemeral by design. Power off, unrack, ship back.
Maintenance and replacement
Under the Turnkey tier, hardware replacement is included. We maintain a spares inventory — drives, memory, NICs, power supplies, USB hubs, complete replacement servers. If a component fails, a replacement ships within 24 hours.
Watch monitors hardware health in real time. SMART data from NVMe drives, ECC error counts from memory, link state from NICs, temperature readings from BMC. Degradation is flagged before it becomes failure. Proactive replacement means less downtime.
Failed drives are securely wiped before return. All data on the buffer is ephemeral — once uploaded and verified, it is deleted. But we treat every drive as if it contains sensitive data, because at some point during its life, it did.
How it fits together
Hardware is the foundation layer. Core runs on it. Ops manages it. Watch monitors it. Direct Line carries the data off it. Every module in the platform touches this hardware.
Ready to talk hardware?
Tell us about your data volumes, device types, and throughput requirements. We will spec a configuration matched to your operation.