KSERVER-1500 1U Server

Kaleidescape’s KSERVER-1500 put a networked movie library into one rack unit. By 2014, sustaining that platform meant working around aging components while preserving the physical interfaces and manufacturing knowledge already in use. The work crossed mechanical definition, power-supply qualification, storage qualification and motherboard replacement planning.

My part of that record sits in mechanical sustaining support. I supplied 3D data for the server, and Rob Wudrick brought me a specific mismatch between the lid’s rack-ear alignment feature and the parts being built. Electrical and disk qualification belonged to the hardware team; purchasing and supplier-transition decisions involved manufacturing.

A server built around removable storage

The 1U enclosure held four front-access disk cartridges. Kaleidescape’s RAID-K architecture allowed a failed cartridge to be replaced while the system continued operating. Storage and playback were separate jobs: the server delivered content over the network to Kaleidescape players elsewhere in the home.

That product arrangement imposed physical constraints on every replacement component. The cartridges shared a shallow enclosure with power, compute and cooling. The published installation guidance required clear space at both the front and rear. Keeping the outside form did not make a new internal component a drop-in replacement.

An earlier cartridge tooling problem

An earlier cartridge problem exposed another interface in the same 1U product family. In September 2008, Michael Clader asked me to reduce the decorative rear bezel on a 3.5-inch drive cartridge. He reported that the change to stamped-metal parts had left the plastic catching on the drive cage and breaking corners. Andy Tripp had observed cracked cartridge plastic while removing a disk from a CEDIA 1U server; his report kept that problem separate from a potentially unrelated backplane incident.

I asked who held the tooling, whether the 3D files were available, how much reduction was needed and whether the team was prepared to retool. Clader identified Steman as the supplier and pointed to Argyle and Arena for the CAD. He proposed a small reduction for me to verify, and suggested adding metal and retexturing the tool before replacing it. The exchange records the problem, the information needed to act and a proposed tooling approach; it does not show an implemented change or a closed field fix.

Keeping the drawing and the built part aligned

The July 2014 handoff began with a file-recovery problem. On July 10, Rob reported that the 1U chassis Pro/ENGINEER part files in Arena were corrupted and asked whether I had a dump of Burke’s chassis files. I said I should have them and would look. That exchange records the requested recovery and my response, without establishing that the Arena files themselves were repaired.

In July 2014, Rob asked me to update the 3D model’s rack-ear alignment detail to match the lid being produced. The model showed a joined extrusion; the production detail differed. He supplied redlined drawings and planned to capture the changes in the two-dimensional documentation for Sanmina. The immediate task was to make the product definition describe the hardware accurately.

This was an existing design with a long manufacturing history. The same base and cover redline revisions appear in correspondence from September 2008, when Rob asked Gabe Cohn at Argyle Design to incorporate production changes. Those records distinguish maintaining the server from creating its original enclosure.

The difference mattered again during a January 2015 supplier review. Prompt Precision was building from redlines that were not incorporated in the 3D files. I confirmed that I had no additional changes queued against the parts. Rob explained that the old redlines predated my involvement and recommended keeping the metalwork at Prompt until the anticipated motherboard or power-supply changes. Elton Armstrong then confirmed Kevin Smyth’s agreement to retain the supplier. Moving the work would also have required recovering its actual manufacturing definition.

A replacement power supply is a system decision

Andrzej Kocan’s February 2015 reports show why power-supply selection remained an integration problem. An SPI FSP150-50LM sample had a touch-fit condition against the lid. An FSP150-5G01 candidate had gone end of life. A Skynet SNP-TX12 sample presented a 20-pin connector where the server needed 24 pins. Enclosed Aras and Sparkle samples differed in fan noise, with the Sparkle SPI300F4BB reported as quiet.

Electrical testing added another constraint. The following week’s report identified a cross-regulation problem with the Skynet sample and questioned whether its 12 V rail had enough current capacity. Noise was one of several qualification criteria still being worked through.

The team also considered using remaining chassis stock after a replacement supply was qualified. Its February meeting notes made that route conditional: a bracket would be needed to fit the new supply to the old chassis. The packaging work had to follow the selected electrical solution and the actual production inventory.

Choosing the next motherboard

The compute search depended on software compatibility, SATA connectivity and power consumption. January 2015 notes considered Baytrail and Celeron options, the availability of four SATA ports and eCos boot compatibility. Greg Curry favored Baytrail’s low-power potential while the meeting notes preferred looking at Celeron to avoid eCos issues with the Baytrail UEFI shell. These were alternatives under evaluation.

The same notes projected a June 2015 runout of 1U inventory. By February 2, the team recorded a commitment to 300 additional 1U chipsets. That was the planned schedule bridge while the replacement motherboard and power supply remained under evaluation.

Qualifying drives in the enclosure

Storage qualification followed its own sequence. In April 2014, Andrzej reported successful testing of a 4TB HGST Deskstar drive in the 1U and 3U servers, including cumulative environmental and disk-stress testing with no recorded errors or retries. He also preserved the unresolved reliability tradeoff. The 1U held four drives horizontally; the 3U held as many as fourteen vertically, with a different exposure to rotational vibration. He proposed longer testing for the 3U rather than treating one result as universal.

In his October 14 report for the previous week, Andrzej recorded release of the 4TB Deskstar NAS under ECO-001442. This was a separate qualification and release from the plain Deskstar tested earlier in the year.

Western Digital alternatives followed a different path. The December 22 meeting recorded WD Red 4TB qualification complete for the 1U, while approving the purchase of WD Red 6TB drives for a long-term test in a four-drive 1U server. The same minutes said WD Red Pro ran too hot in the 1U and 3U and could not be qualified.

By January 5, the meeting record described WD Red 6TB upgrade-drive qualification as complete. Andrzej and Isaiah were setting up long-term reliability testing, with release planned by the end of the week for the Extranet price list. That report advanced the WD branch beyond the December plan while leaving actual release and long-term results unestablished. It was separate from the later thermal result for 6TB HGST Deskstar NAS drives.

The next capacity step again met the enclosure’s limits. In February 2015, Andrzej’s review of Isaiah Erb’s 6TB Deskstar NAS testing reported temperatures about 5–6 degrees Celsius too hot in the 1U server. The following week’s report reviewed the thermal disk-stress results and again found the drive about 5 degrees Celsius too hot.

What the record establishes

The engineering work was a set of connected decisions about an existing product: recover the built mechanical definition, retain a supplier that understood it, evaluate replacement components in the actual enclosure, and make schedule commitments while qualification continued. My documented mechanical involvement sits within that cross-functional effort.

By the latest reports, the 4TB NAS drive was released and the team had agreed to keep legacy metalwork at Prompt. The replacement power supply and motherboard were still under evaluation. The final 1U refresh configuration and shipment outcome remain unresolved in the reviewed records.