Publications
The “Data infrastructure” series: we unpack the engineering trade-offs where design decisions most often go wrong.
Debt and Default
Infrastructure requirements have grown faster than hardware depreciates. Five real cases — from a fleet built in advance to infrastructure pieced together from whatever budget was available.
Read the issueLayer by layer
Continuing the conversation about kilowatts. Last time we calculated from specifications — from what companies were about to buy. This time we calculate from what's already installed: two sites, five generations of equipment, engineered systems with their own rack, and two temperature investigations — in one of which the obvious explanation turned out to be wrong.
Read the issueThe rack runs out early
Infrastructure is purchased in cores and terabytes, but deployed in kilowatts. Power calculations for configurations from real projects, the arithmetic of phases and feeds, and a 45-node complex where 71 percent of the electricity bill is paid just for being switched on.
Read the issueThe fork after vSphere
Five replacement options, two starting points — the classic array-based setup and a vSAN cluster — and what happens to storage and networking once the 'ESXi plus Fibre Channel array' combo disappears from the picture.
Read the issueLines and cores
A customer picks an application system and sees one line in the budget. They end up paying for four. A case study: a sizing dispute that nearly quadrupled a three-year licensing subscription, and the measurement that settled it eighteen months later.
Read the issueFour circuits
Renting versus owning in virtualization, backup, analytics, and AI workloads. Four circuits, and in each the line falls differently — almost never where people draw it.
Read the issueFour ages
How much support to buy up front, why keep spare parts under an active contract, how to tell that hardware can no longer be trusted one hundred percent, and when replacement turns out cheaper than renewal.
Read the issueThree substitutions
How a backup system is actually sized: how the protected perimeter differs from storage capacity, why the vendor's stated data reduction ratio doesn't hold on an encrypted stream, and what to ask a vendor before the budget is approved.
Read the issueTwo relocations, two strategies
A database cluster moved between two server rooms, and an office consolidated onto two racks — in one week. Why the plan hinged on neither case on rigging, and what to ask before you set a date.
Read the issueBackup: counting the wrong things
Capacity, retention depth, backup window — and not a word about how long it takes to bring a service back. A breakdown of the forks where backup system sizing parts ways with reality.
Read the issueA tape library for those who didn't design it
How this layer of a backup system is built, what breaks in it, why LTO-10 changes the planning arithmetic, and where the cartridge count actually comes from.
Read the issueHardware for BigData
Where the array belongs, where local disks belong, and where virtualization shouldn't happen at all. A breakdown of three forks that shape the architecture and budget of a data platform.
Read the issue