KSERVER-5000

Kaleidescape’s KSERVER-5000 put fourteen disk cartridges behind a curved front door in a 3U movie server. Opening the door exposed the cartridges without removing the top cover. The product combined a recognizable domestic face with the cooling, access and manufacturing constraints of a densely packed storage chassis.

My documented contribution was mechanical sustaining: reviewing a cooler alternative, creating the 3D model and reformatted drawing for a replacement power-supply adapter, and helping manufacturing interpret that adapter’s fabrication details. The work joined an existing platform. Rob Wudrick and Argyle’s earlier engineering remained part of its design history.

Fourteen drives behind the front door

The photographs show the package’s two different faces. Closed, the broad front panel concealed the cartridge bank. Open, its perforated inner panel lowered below a row of fourteen vertical drives, carried by linkages at either side. The hinge detail makes the connection between the moving door and the chassis visible. An archived section drawing separates the screen, glow PCB, inner panel and chassis into distinct layers. These were service and assembly interfaces as well as elements of the product’s appearance.

This architecture was already established before my arrival at Kaleidescape. The base assembly’s BOM names Rob Wudrick, while the drive cage, guides, hinge plates, inner door and fan-tray records carry Argyle ownership and 2006 dates. My work on later changes operated within that inherited arrangement. The selected examples here come from 2008, 2014 and early 2015.

A cooler choice constrained by the schedule

In July 2008, hardware engineer Nathan Forster had found a passive aluminum CPU heatsink with acceptable performance in his evaluation. Its sample was resting on the CPU under gravity. He asked me to create an internal extrusion drawing and a suitable attachment method; the surrounding case could not be reused because it obstructed airflow. I asked for the drawings so we could discuss the work.

The remaining work extended beyond drawing an extrusion. It needed mounting design, prototypes, qualification and a production order. Richard Lane directed the team toward an off-the-shelf cooler: selecting a solution in August and implementing it in September left insufficient time for a custom extrusion. A proprietary design could be reconsidered after the available alternatives had been evaluated.

The team continued with active-cooler evaluation. In August, Forster compared the server with the replacement CPU fan switched off and on using an external supply. Rob Wudrick and Andrzej Kocan also assessed the audible change. Their observations described the additional fan sound blending into the existing server noise. That test belonged to the active replacement cooler; the earlier passive extrusion remained a separate proposal.

Making the power-supply adapter explicit

A later power-supply change required an adapter at the rear of the server. Drawing 501-1211-00 identifies it as the IEC320 filler for the Sparkle FSP650. Its revision history records Rob Wudrick’s initial rough 2D drawing in February 2014, followed by my creation of the 3D model and reformatted print on March 6.

The assembly view describes how the sheet-metal pieces come together: align them with half-shears, then weld around the perimeter. A separate view calls for the PEM nuts. These details serve different purposes. The half-shears locate pieces during fabrication; the completed adapter still has to provide the correct dimensions and mounting locations.

Checking the requirement behind an apparent defect

That distinction became consequential during incoming inspection in May 2014. Sanmina reported a batch of 100 adapters with two apparently missing holes. Manufacturing escalated the discrepancy as a risk to KSERVER-5000 production.

I explained that the 0.125-inch features were half-shears used for alignment during fabrication. They had no further function once fabrication was complete. My proposed disposition depended on the finished part’s physical dimensions and PEM locations being correct, and I asked the team to install a part to confirm the fit.

The clarification established what needed to be checked, but it did not complete the check. The power supply had not yet arrived, so Sanmina could not assemble the sample that day. Yamil Ramirez also requested an ECO or DCO update so future inspections would not reject the same condition. My later status reports and exchange with Wudrick still carried the documentation question, including whether the earlier adapter needed similar treatment. The retained record does not establish the final fit or shipment outcome.

Qualification remained a team effort

The electrical qualification had its own evidence and ownership. On May 2, Andrzej Kocan reported that the server with the FSP650EGN-KS supply had passed radiated and conducted emissions pre-scan at Ultratech. That result supported the hardware team’s PSU work; it did not answer the later mechanical installation question.

Drive changes brought another set of constraints. Kocan’s April evaluation distinguished the fourteen vertical drives in the 3U server from the smaller horizontal arrangement in the 1U product, and called for longer testing of the proposed drives.

The intervening reports show why a drive test and the interpretation of its health counters were separate questions. In early June, Matthew Frazer reported clean stress-test logs alongside changes in SMART values gathered by a third-party tool. Nathan Forster asked what those changes meant, and Kocan offered to seek clarification from the tool supplier and HGST. The team added HGST’s HiTest to its checks. By mid-June, HiTest reported passes while the separate Landscape monitor still carried an earlier SMART warning.

On June 25, Mark McKenzie authorized release of the Deskstar/3U change and reported that operations had stopped buying the previous drives, with rollout planned for early July. The following morning, Wudrick identified ECO-001427 as moving through the system. Testing continued: Patel’s October report included a fourth 3U server testing Deskstar NAS drives. These were distinct steps in the component transition, without the reports establishing a delivered rollout or resolving every monitoring discrepancy.

In December, Swapnil Patel’s long-term test summary reported clean read/write stress logs alongside SMART errors seen through a separate monitoring system. The mixed observations required investigation rather than a blanket reliability claim.

The drive work continued beyond the initial Deskstar ECO. September minutes recorded completed 3U voltage-ripple testing for Deskstar NAS, followed by a staged ECO and an older 3U chassis added to long-term testing. Kocan’s report for the week ending October 10 recorded release of the HGST 4TB Deskstar NAS component under ECO-001442. The December 22 hardware meeting drew a separate boundary around Western Digital alternatives: it recorded WD Red 4TB qualification complete for the 3U but no plan to ship 3U servers with those drives, with long-term testing still to be done. WD Red Pro ran too hot for qualification in the 1U and 3U.

Substituting the drive indicators

The sustaining work also reached the drive indicators. In June 2014, Andrzej Kocan and Matthew Frazer evaluated replacement blue LEDs after the existing part became difficult to source. Avago samples were too bright. Frazer found that the LiteOn samples worked at very low current, although LiteOn did not endorse that operating condition; an Osram part performed well on the backplane but might be too weak on the pogo board. By the week ending June 27, Kocan reported OSRAM and LiteOn LEDs released for the drive indicators and 3U power button under ECO-1428.

Keeping drawings aligned with production

The contract-manufacturer transition required preserving known parts and processes. By September 29, the hardware minutes called for moving server assembly to the new manufacturer while keeping 1U/3U metalwork at Prompt and the existing SATA-board fabricators. The plan separated the assembly transfer from changes to parts that had taken time to get right.

The sustaining record also shows how an established supplier can hold details that the current CAD package has not absorbed. In January 2015, a drawing update effort found legacy redlines on server sheet metal. I reported no additional changes queued against the listed parts. Wudrick explained that the redlines predated me and were still absent from the current 3D drawings; Elton Armstrong subsequently reported the decision to keep the 1U and 3U metalwork with Prompt.

Later team correspondence shows why inspection findings also needed context. In February 2015, Nathan Forster reported cosmetic defects in eight of 32 front-bezel assemblies on hand, but cautioned that the stock’s age and transfer history were uncertain. That sample could not establish a production yield. Steman’s Weddy Wang reported no matching defect in 120 parts checked from 500 ready to ship to Hunter and added the defect to the inspection list. The exchange records a supplier check and a future inspection step, without establishing a molding cause or a completed shipment.

The recurring engineering task was to make the design intent legible at the interface between model, drawing, supplier and assembled product. The PSU adapter provides the clearest example: a feature used to make the part had been interpreted as a requirement of the finished part. Resolving that interpretation required both a mechanical explanation and an explicit path to fit verification and document control.