Large-Scale Imaging Server
Implementation — Microtek

The Situation

Microtek is a Colorado-based education technology firm that has built and deployed custom PC images for schools, districts, and organizations for over 20 years. Their operation runs as a purpose-built assembly environment — institutions order devices, Microtek images them to spec, and ships them ready to deploy. Volume is the business model, and the imaging pipeline is the throughput constraint.

At the time of the engagement, their imaging setup was a repurposed workstation running Norton Ghost — a legacy choice that had served them for years but was increasingly the bottleneck for everything downstream. The workstation's local spinning storage was the rate limiter: slow reads meant slow image delivery, which meant technicians were waiting on hardware instead of moving through the line.

This engagement was delivered through Information Systems Integrators (ISI), where I was responsible for the technical implementation from design through handover.

The Challenge

• The imaging "server" was a repurposed workstation — not built for sustained parallel read workloads at production volume
• Norton Ghost's legacy architecture wasn't designed for modern PXE-based network deployment at scale
• Slow local spinning storage meant image delivery was the bottleneck — technicians waited on the machine instead of the machine waiting on technicians
• No true server-grade network infrastructure: the workstation's single standard NIC couldn't feed multiple simultaneous deployments without degrading each one
• Target endpoints — HP and Lenovo laptops selected by educational institutions — shipped with mSATA drives capable of write speeds the old setup couldn't come close to saturating
• Scaling throughput meant adding technicians, not improving infrastructure — a cost structure that didn't fit the business

The Approach

The engineering goal was to make the imaging server fast enough that the endpoint drive became the limiting factor — not the network, not the server storage, and definitely not the server's ability to serve multiple machines simultaneously. If we could saturate the mSATA write speed on a laptop over a 1G connection, the deployment time would be as short as physics allowed. Everything else was about removing obstacles between the image and the drive.

• Server platform: Replaced the workstation with a purpose-built 1U rack server with a storage subsystem selected specifically for high-throughput parallel reads. The server needed to serve multiple simultaneous imaging sessions without any single session degrading — which meant the local storage had to handle intensive concurrent read I/O without becoming the bottleneck when the line was running at full capacity.
• 10G network uplinks: The server connected to two network switches via 10GbE. This gave the imaging server 10x the uplink bandwidth of any individual endpoint — meaning it could feed ten simultaneous 1G imaging sessions at full line rate before the server's own network became a constraint. In practice, the deployment floor ran well within that headroom.
• 1G to endpoints: Each HP and Lenovo laptop on the assembly stations connected to the switches over standard 1GbE. The critical finding during testing: the mSATA drives in the target laptops could be saturated via the 1G connection — meaning the image was arriving faster than the drive could write it, and the drive write speed was the actual ceiling. This is what you want. The network was no longer the slowest thing in the chain.
• PXE boot and Macrium Reflect Deployment Toolkit: Configured DHCP and TFTP to serve the WinPE boot environment over the network. Machines booted directly into the Macrium Reflect imaging environment with no local media involved. The Deployment Toolkit managed the golden image library, hardware-specific deployment profiles for HP and Lenovo target SKUs, and the technician-facing menu that kept the workflow simple on the floor.
• Assembly-line workflow design: The real unlock wasn't just speed — it was what speed enabled. With deployment time short enough, the workflow became a true assembly line: a technician connects a laptop, initiates the PXE boot, selects the correct golden image from the menu, and moves to the next station. By the time they've worked down the line, the first machine is done and ready for QA and boxing. Technician time and machine time fully overlapped. No waiting.
• Golden image management: Built out separate deployment profiles for the HP and Lenovo laptop lines in active use across Microtek's institutional clients. Established a process for capturing, versioning, and testing updated images before promoting them to the production deployment menu — so a bad image couldn't silently ship to a client order.