Okay, so I have this Cobalt Qube 3 sitting in my home office for months on end, and as much as I love the little box, I really want to repurpose it (which might/might not include hollowing out the hardware and only retaining the chassis). First, a look at the chassis:
For those youngins' out there who are too young to know, or for those who have short memory retention, a Cobalt Qube 3 is a piece of Microserver hardware that was released back in the old dotcom boom of the early 2000s. Cobalt made a bunch of decent network appliances based on MIPS / PMC-Sierra R5000 chips, then they switched over to AMD K6 (K6-2 Socket 7 Notebook CPUs), and then finally Intel Pentium III Coppermines. In late 2002 they were bought out by Sun Microsystems, who subsequently shut them down and liquidated their assets (another competitor gone). The Qube3 was the last released product in the Qube microserver line, while its 1U small form factor Raq continued until the final product (The Raq XTR) came out back in 2002.
So, what's in the back? A 2 line LCD display, 2 10/100 ethernet ports (NatSemi MacPhyter based) a serial port, an optional SCSI port and a single full-height 32 Bit PCI (not PCIe) slot. The ventilation is via a noisy little 60 mm fan, and the power brick is proprietary.
So what's inside? There is a TOP LOADING chassis holding 2 3.5" EIDE drives (obviously not hot swappable) with its own molex power connectors (note that the original EIDE cable does not run tio the back like that). There is a custom motherboard with a single AMD K-6-2 CPU and 2 PC100/133 RAM slots. You can see on that photo the SileX deep-chord/slow RPM fan with silcone mounting screws that I installed to minimize noise.
So here's a photo of the motherboard - you can see the SCSI connector above the coin battery, which is directly under the CPU/heatsink. The original K6-2 450MHz has been replaced with a K6-3+ 450MHz. You can see the full size PC100 DIMM units - there are 2 sticks of 512MB DIMMs populated on the slots. You can see the ALi M1541 Socket 7 Northbridge on the board along with the LSI SCSI controller next to the slot. On the bottom next to the EIDE port is the ALi M1543 southbridge. The 2 large chips are the NatSemi McPhyter NICs, the NIC magnetics and the PHY. The chip with the sticker is the 1MByte EEPROM chip holding the boot image.
Here's a closer look at that bottom board - the ALi Southbridge is on the lower left, the 2 NatSemi network controllers are arrayed along the middle, the magnetics and PHYs are to its right. The connector above the PHYs is for the LCD panel.
Here's the top board. The CPU, the LSI 53C825 SCSI controller, the SCSI port (goes to an HDBP50 connector outside the chassis via a ribbon cable, the Northbridge chip, the RAM slots and the CPU/heatsink are visible.
An even closer look at that top board.
So, here's how it was wired up until the power-down a few months ago - I had a Y-cable split on the power adapter where one of the IDE cables drive the primary disk (was an 80GB Samsung Spinpoint M7 laptop HDD along with a 2.5" to 3.5" HDD adapter, nowadays connecting an 8GB EIDE DOM from an HP gt7725 thin client).
What about the other 2 drives? Well, the power is fed off that molex connector pair, while the disks will be connected via a Promise Ultra133 TX2 PCI adapter (although I might've replaced it with a SATA card not too long ago).
Of course, for those who are wondering what is inside that blue chassis, here's what it looks like after it's all emptied out. You can see the fan mount, the SCSI port connector and the dual green LED lamps on the chassis.
So, that covers the hardware, which consists of:
1x AMD K6-3+ @ 450MHz
2 x 512MB PC100/133 DIMM modules
ALi M1541/1543 Northbridge/Southbridge
2 IDE ports
1 8GB EIDE DOM
1 2x16 segment LCD panels with buttons
1 LSI SCSI-2 port
1 Serial port
2 NatSemi McPhyter 10/100 NICs
1 USB 1.1 port
1 32 Bit PCI slot, currently used to house:
1 Promise Ultra133 TX2 adapter, which is used to connect -
2 2TB Western Digital EIDE drives (both of which shows serious signs of eminent failure)
So that covers the hardware. Now in terms of software, the Cobalt Qube/Raq series (except the XTR) all use a Linux kernel based boot ROM and does not have any recognizable BIOS to speak of. The boot ROM image is loaded as a stage1. Then what you have to do is compile a Linux stage2 kernel on its hard drive (which I remember has to be less than 1800k on a bz2, so strip that kernel, please). Userland on my Qube was Debian Wheezy with a custom compiled 3.4.8 kernel (someone broke the touch panel code on the 3.5.x kernels, and I didn't have any free time to kick kernel 4 screaming into my tortured little box). To keep within the spirit of the Cobalt stack with its non-threatening GUI based management, it rent webmin on top of Debian 7.
So where does that leave you, dear readers?
Well, It has been 5 years since I bought my MicroServer G7 (which I consider to be the Qube 3's spiritual successor), and even then it has been relegated from my main server to being my iSCSI box (the computing has been offloaded to a more powerful thin client acting as a hypervisor). So it's time to figure out what I should do with this 18 year old box (if it's human it would be allowed to go clubbing).
So, I can do the following:
a) Hollow out its guts and replace it with new components...?
But what? I want to hear from you guys.
The only rule that I have is that it cannot be something ridiculous like a stack of RaspBerry Pi 3s or some fly-by-night ARM boards with questionable hardware integration support (because lord I learned that one buying the NTC C.H.I.P. - that was not even mainline Debian Linux). I expect it to do native SATA and have ability to do some clever data protection stuff, I mean, the original is 18 years old and support RAID1 natively, dammit. If it's not doing something like zraid2 or btrfs pools, it's gonna get tossed out.
b) Keep it running as-is.
One of the funnier things about Linux is how you can run it on a 1024 node GPGPU cluster, and also on a freaking WiFi enabled toaster. Well, the K6-3+ is pretty much going to be outgunned by a Pi2 running a toaster. The question is...how far can you stretch the ridiculousness? Can it run kernel 4.12+ and a modern userland (probably Slackware instead of Debian)? Does it need a Spectre patch? How far can we push it? If we do, what should we torture the little box to do?
Before you laugh, don't forget that DD-WRT still have modern/up-to-date builds for the original WRT54G, and that thing has, what, 16MB of RAM and 4MB flash...never underestimate the power of a determined IT guy with nearly infinite patience.
c) Viking burial.
If we are gonna send the little blue guy off, might as well cram it full of Iomega Zip Drives (itself containing a bunch of Apple ][e disk images and Commodore Basic program listings) and send it off to Best Buy. At least the look on that poor worker there will be worth a laugh or 2.
For those youngins' out there who are too young to know, or for those who have short memory retention, a Cobalt Qube 3 is a piece of Microserver hardware that was released back in the old dotcom boom of the early 2000s. Cobalt made a bunch of decent network appliances based on MIPS / PMC-Sierra R5000 chips, then they switched over to AMD K6 (K6-2 Socket 7 Notebook CPUs), and then finally Intel Pentium III Coppermines. In late 2002 they were bought out by Sun Microsystems, who subsequently shut them down and liquidated their assets (another competitor gone). The Qube3 was the last released product in the Qube microserver line, while its 1U small form factor Raq continued until the final product (The Raq XTR) came out back in 2002.
So, what's in the back? A 2 line LCD display, 2 10/100 ethernet ports (NatSemi MacPhyter based) a serial port, an optional SCSI port and a single full-height 32 Bit PCI (not PCIe) slot. The ventilation is via a noisy little 60 mm fan, and the power brick is proprietary.
So what's inside? There is a TOP LOADING chassis holding 2 3.5" EIDE drives (obviously not hot swappable) with its own molex power connectors (note that the original EIDE cable does not run tio the back like that). There is a custom motherboard with a single AMD K-6-2 CPU and 2 PC100/133 RAM slots. You can see on that photo the SileX deep-chord/slow RPM fan with silcone mounting screws that I installed to minimize noise.
So here's a photo of the motherboard - you can see the SCSI connector above the coin battery, which is directly under the CPU/heatsink. The original K6-2 450MHz has been replaced with a K6-3+ 450MHz. You can see the full size PC100 DIMM units - there are 2 sticks of 512MB DIMMs populated on the slots. You can see the ALi M1541 Socket 7 Northbridge on the board along with the LSI SCSI controller next to the slot. On the bottom next to the EIDE port is the ALi M1543 southbridge. The 2 large chips are the NatSemi McPhyter NICs, the NIC magnetics and the PHY. The chip with the sticker is the 1MByte EEPROM chip holding the boot image.
Here's a closer look at that bottom board - the ALi Southbridge is on the lower left, the 2 NatSemi network controllers are arrayed along the middle, the magnetics and PHYs are to its right. The connector above the PHYs is for the LCD panel.
Here's the top board. The CPU, the LSI 53C825 SCSI controller, the SCSI port (goes to an HDBP50 connector outside the chassis via a ribbon cable, the Northbridge chip, the RAM slots and the CPU/heatsink are visible.
An even closer look at that top board.
So, here's how it was wired up until the power-down a few months ago - I had a Y-cable split on the power adapter where one of the IDE cables drive the primary disk (was an 80GB Samsung Spinpoint M7 laptop HDD along with a 2.5" to 3.5" HDD adapter, nowadays connecting an 8GB EIDE DOM from an HP gt7725 thin client).
What about the other 2 drives? Well, the power is fed off that molex connector pair, while the disks will be connected via a Promise Ultra133 TX2 PCI adapter (although I might've replaced it with a SATA card not too long ago).
Of course, for those who are wondering what is inside that blue chassis, here's what it looks like after it's all emptied out. You can see the fan mount, the SCSI port connector and the dual green LED lamps on the chassis.
So, that covers the hardware, which consists of:
1x AMD K6-3+ @ 450MHz
2 x 512MB PC100/133 DIMM modules
ALi M1541/1543 Northbridge/Southbridge
2 IDE ports
1 8GB EIDE DOM
1 2x16 segment LCD panels with buttons
1 LSI SCSI-2 port
1 Serial port
2 NatSemi McPhyter 10/100 NICs
1 USB 1.1 port
1 32 Bit PCI slot, currently used to house:
1 Promise Ultra133 TX2 adapter, which is used to connect -
2 2TB Western Digital EIDE drives (both of which shows serious signs of eminent failure)
So that covers the hardware. Now in terms of software, the Cobalt Qube/Raq series (except the XTR) all use a Linux kernel based boot ROM and does not have any recognizable BIOS to speak of. The boot ROM image is loaded as a stage1. Then what you have to do is compile a Linux stage2 kernel on its hard drive (which I remember has to be less than 1800k on a bz2, so strip that kernel, please). Userland on my Qube was Debian Wheezy with a custom compiled 3.4.8 kernel (someone broke the touch panel code on the 3.5.x kernels, and I didn't have any free time to kick kernel 4 screaming into my tortured little box). To keep within the spirit of the Cobalt stack with its non-threatening GUI based management, it rent webmin on top of Debian 7.
So where does that leave you, dear readers?
Well, It has been 5 years since I bought my MicroServer G7 (which I consider to be the Qube 3's spiritual successor), and even then it has been relegated from my main server to being my iSCSI box (the computing has been offloaded to a more powerful thin client acting as a hypervisor). So it's time to figure out what I should do with this 18 year old box (if it's human it would be allowed to go clubbing).
So, I can do the following:
a) Hollow out its guts and replace it with new components...?
But what? I want to hear from you guys.
The only rule that I have is that it cannot be something ridiculous like a stack of RaspBerry Pi 3s or some fly-by-night ARM boards with questionable hardware integration support (because lord I learned that one buying the NTC C.H.I.P. - that was not even mainline Debian Linux). I expect it to do native SATA and have ability to do some clever data protection stuff, I mean, the original is 18 years old and support RAID1 natively, dammit. If it's not doing something like zraid2 or btrfs pools, it's gonna get tossed out.
b) Keep it running as-is.
One of the funnier things about Linux is how you can run it on a 1024 node GPGPU cluster, and also on a freaking WiFi enabled toaster. Well, the K6-3+ is pretty much going to be outgunned by a Pi2 running a toaster. The question is...how far can you stretch the ridiculousness? Can it run kernel 4.12+ and a modern userland (probably Slackware instead of Debian)? Does it need a Spectre patch? How far can we push it? If we do, what should we torture the little box to do?
Before you laugh, don't forget that DD-WRT still have modern/up-to-date builds for the original WRT54G, and that thing has, what, 16MB of RAM and 4MB flash...never underestimate the power of a determined IT guy with nearly infinite patience.
c) Viking burial.
If we are gonna send the little blue guy off, might as well cram it full of Iomega Zip Drives (itself containing a bunch of Apple ][e disk images and Commodore Basic program listings) and send it off to Best Buy. At least the look on that poor worker there will be worth a laugh or 2.
Last edited: