Return to Floodgap ANSwers

Floodgap ANSwers: The WGS 9150 and the Story of Wormhole

The Apple Network Server line would not have existed without the Apple Workgroup Server 9150, the machine that as Green Giant forced Apple into a new server design direction. And it just so happens I own one.

If you'd like to use these pictures somewhere else, please ask me first. Thanks!

The Story of Wormhole, Green Giant and Starbucks

Apple had a long history with abortive RISC projects, starting as far back as 1986 with the resource-intensive Aquarius. Designed to create a four-core RISC processor of Apple's own to succeed the Motorola 68000 series, the over fifty engineers assigned to the project could not develop a workable prototype. CEO John Sculley eventually cancelled Aquarius and the Cray supercomputer he bought to design it and recruited expert designer Hugh Martin in 1989 to actually develop a computer instead, using an existing CPU. The new Jaguar team adopted the Motorola 88000 "88K" architecture and began creating a design with four 88100 cores with the intention of developing a successor to the Macintosh. A renegade group led by veteran Apple engineer Jack McHenry objected, saying a completely new and incompatible computer would doom Apple, and established the competing Cognac project (a wordplay on RISC pioneer John Hennessey) in 1990.

Cognac succeeded where Jaguar failed by simply succeeding. (Jaguar later became the similarly vapourware Tesseract project, and subsequently folded, having never actually built any working hardware. The codename of the Power Macintosh 7500, "TNT," actually stands for "The New Tesseract" -- ironically a descendant of Cognac.) Using a heavily modified Macintosh LC, the Cognac team quickly constructed a 88100-based prototype (the "RISC LC" or "RLC") that to the great shock of Apple management ran existing Macintosh software with full compatibility. The key was Gary Davidian's high performance 68K emulator, which enabled even the operating system itself to run in emulation under a low-level nanokernel (today we would call this an early type of hypervisor). The emulation layer was also quite portable, which was a substantial advantage, because the Cognac project had also previously existed in one-off MIPS and ARM-based prototypes as well. When Apple, IBM and Motorola established the AIM alliance in 1991, because Apple now already had a working hardware design they successfully persuaded IBM to implement the same 88K bus on the new PowerPC. Since the hardware was thus able to remain largely the same, the RLC was easily adapted to the new architecture by simply rewriting the emulator one final time. Because Jaguar/Tesseract failed to deliver, it was the Cognac-based RLC that evolved into Piltdown Man, which became the Power Macintosh 6100 in March 1994.

Recall from our ANS history that by this time Sculley had created the Server Group and tasked it with developing an enterprise-grade platform. Their first and by far most popular model was the Apple Workgroup Server 95, released in March 1993, based on the workhorse Quadra 950. Although the AWS 95 was almost identical to the Q950, the extra-capacity drive carrier, PDS fast-SCSI card and tape drive it was bundled with, plus the fact it was the most expandable and powerful machine then capable of running A/UX, made it substantially more popular with enterprise Mac shops than the Workgroup Server 60 (Centris 610) and WGS 80 (Q800) released later that year.

RLC was amazing, but it could not run A/UX, and it wasn't ever meant to; the emulation overhead would have made it unacceptably slow for high-performance applications. Instead, Apple planned to rewrite it for the PowerPC by merging it with AIX based on the Mach-based OSF/1, where both A/UX and AIX would be personalities on this new joint OS, and the AIM alliance Taligent project was supposed to create the successor to Mac OS by layering on top of the old System 7-era Pink concept (Copland). This should have, at least in theory, covered both bases, but by the end of 1993 neither initiative had anything even partially working. (By the time the Apple Network Servers were released in 1996, while the NuKernel-based Copland prototype did finally exist in a rudimentary, bug-infested developer build it could not run anything of substance, let alone Taligent's CommonPoint application environment. The OSF/1 A/UX never even made a formal release. However, some of the low-level parts were probably recycled for the Mach underpinnings of MkLinux, which in turn were recycled for OS X -- see the images below.)

As MacOS was itself not ideal as a server operating system, other alternatives were considered. NetWare was the star network operating system (pun intended) by the early 1990s, since it ran on relatively inexpensive hardware and was easy to deploy to clients; Novell was even a potential merger suitor of Apple in 1992, and was the instigator of the original Star Trek (Mac-on-Intel) project that indirectly led to Marklar. Significantly, NetWare 3.0 already existed in a machine-agnostic form called Portable NetWare which could be built from source by an OEM and run on an existing OS. This seemed too good to be true: all Apple had to do was rebuild it for the PowerPC and slap it on top of a bootloader, which could even be MacOS-based; the project to make this a reality was variously christened Wormhole and Deep Space Nine. Since the AWS 95 was the most popular of the existing server line, the Server Group recycled the same case, even the key lock, and created a new PowerPC-based 601 motherboard to fit by stealing features from the in-development Power Macintosh 8100. However, the 8100's PDS slot was intended for video cards as well, so Apple had to sacrifice a NuBus slot to get it to fit in the same enclosure, and ended up replacing the top floppy drive bay with a CD-ROM and moving the floppy to the bottom to maintain the two 5.25" bays of the Q950 case. (The 9150 was the only Macintosh ever released in this physical configuration, but it isn't clear if the prototype had both these features, or if the design was modified later in production.) Codenamed Green Giant, this prototype was loaded up with a development build of Wormhole and demonstrated to testers in 1993.

Wormhole's reception was, at best, tepid; testers regarded the choice of NetWare as short-sighted, later proven correct. Apple management hastily promised Unix instead, but eventually redirected resources to the new Shiner project which became the ANS. However, they still had a working unit in Green Giant that was almost ready to ship, so the Server Group simply eliminated the Portable NetWare overlay and it shipped with MacOS like the other Workgroup Servers. Combined with WGS systems based on the Power Macintosh 6100 and 8100 (the WGS 6150 and 8150) as the Starbucks project, the complete line finally reached the market just one month after the original Power Macintosh 6100/7100/8100 lineup, in April 1994.

Despite Wormhole's poor reception, CEO Michael Spindler was still determined to get NetWare onto at least one Apple system anyway; as Novell turned Portable NetWare, based on NetWare 3, into Processor Independent NetWare based on NetWare 4, Spindler promised potential buyers (and the media) at the WGS' introduction that "at least one system [would be] bundled with PIN." For a period of time this was even going to be the Apple Network Servers themselves; this project was called Cyberpunk and actually existed, based on a PowerPC-native alpha build of NetWare 4.11, though the extant bootable CD does not work on the ANS nor on the 9150. Apple cancelled Cyberpunk outright in October 1995 and Novell decided to develop themselves for the CHRP PowerPC systems instead; this port was completed by Cygnus under contract and saw little use until Novell cancelled the entire alternative NetWare concept with NetWare 5 in 1998.

Apple eventually bumped the WGS 9150 to 120MHz in April 1995 ("Zephyr") before discontinuing the Starbucks systems in February 1996 altogether for the new PCI-based Workgroup Servers and, of course, the ANS.

Differences from Starbucks

Although grouped with Starbucks, Green Giant is a distinct system. Besides its unusual drive placement and "five-slot" layout (4 NuBus + 1 PDS), it also retains the separate audio-in phono/RCA jacks of the Q950/AWS 95. Similarly, the 9150 is the only one of the original 601 NuBus systems to not have the HDI-45 video connector; instead, it has a "regular" Mac DB-15. However, it uses the same Ariel II video chip as the 6100/7100/8100 and does not use the DAFB (Direct Access Frame Buffer) chip of the Q950, nor does it have VRAM. In that sense it is more closely related to the RBV (RAM-Based Video) chip of the Mac IIci and IIsi than the other VRAM-based 68K framebuffers.

In addition to the CPU fan, the 120MHz 9150 has an active Peltier-effect thermoelectric cooler, the first of three Apple 601 machines that used it (the first 7200/90 and the 7200/120 were the others) and the only one of the NuBus machines that ever did. The 9150/120 and the 7200/120 were the fastest PowerPC 601 machines Apple shipped, using the second-generation 601+. The WGS 9150 came standard with tape, CD, 512K of L2 cache and two 1GB or 2GB drives depending on configuration, with a software RAID solution, server software, backup software and System 7.1.2 on disc.


brinton

These are images of brinton, my 9150/80.

Front badge.

Entire original image (132.4KB)   

Backplate, with a 1994 copyright date.

Entire original image (135.4KB)   

Portrait. This unit, like the 7100, is "two-tone" with plastic front and side panels mounted on a painted metal case. Although I did a bit of restorative buffing to the front panels to reduce the yellowing ...

Entire original image (86.3KB)   

... it contrasts clearly with the nearly perfectly white painted metal side. The key is the same type of wafer-lock key the Apple Network Server uses, but the keyswitch is substantially different. Instead, it functions identically to the Q950/AWS 95: if turned to the 0 position, it actually powers the server off; if turned to the 1 position, it functions normally; if turned to the lock position, it turns on, stays on, and ignores the floppy drive and activity on ADB. If you turn the switch to 1 from the locked position, ADB is enabled, and the floppy drive is enabled (you can go back and forth). However, if you turn the switch to 0 from 1, the server immediately powers down, and won't power back on if the switch isn't turned back to 1.

As a result, you don't need a power switch. Just turn the key all the way to lock (powers up), and turn it back to 1.

Entire original image (68.8KB)   

Front badge with low-mounted floppy drive and programmer and reset switches.

Entire original image (69.2KB)   

The rear of the unit is identical to the Q950, with the same port layout.

Entire original image (109.6KB)   

The slots use the more robust Q950-style snapouts rather than the flimsy ones on the other NuBus Macs. From the rear they appear all to be NuBus, but they aren't.

Entire original image (88.5KB)   

From top to bottom: built-in Ariel video (on a DB-15), AAUI Ethernet, SCSI, serial and printer ports, ADB, phono audio out, and 1/8" stereo audio in and out. The phono jacks and 1/8" output are the same audio source. This unit has an Apple Display Card 8*24 (670) NuBus video card in the second position, which is the topmost NuBus slot.

Entire original image (95KB)   

The side door, which is held in by rather fragile-appearing tension clips and a screw, with the card guides.

Entire original image (139KB)   

The WGS 9150 and the AWS 95 both have the full five-disk carrier, which is on top, and the 9150 has the DDS-DC tape drive as standard equipment. The carrier has three top slots and two lower slots, the bottom of which has two SCSI ID cables you can set on the carrier itself.

Entire original image (126.6KB)   

By removing two screws and pulling on the rope-cable handle, the carrier is removed from the inside. The power supply sits in the middle. The brown chassis (steel, incredibly heavy, totally impervious) and Velcro cable guides also hail from the Q950.

Entire original image (121KB)   

The tape drive/CD-ROM cage comes out with a separate screw, and then the entire mounting plate on top of the power supply slides back.

Entire original image (137.4KB)   

The SCSI cable is also unusual. It is labeled for which goes to which slot on the carrier and in the front drive bays, which is important because there is only one SCSI port on the logic board. It has a terminator.

Entire original image (111.1KB)   

The visible mainboard beneath the power supply with the front panels off.

Entire original image (135.4KB)   

The DRAM slots and soldered RAM. All units have 8MB soldered on; the /80 came with just that, but the /120 had 8MB worth of SIMMs as well for 16MB standard. Up to 256MB RAM can be installed with 80ns+ 72-pin SIMMs for a total of 264MB.

Entire original image (147.3KB)   

The slots themselves. Notice there are only four; the top slot is a PDS slot.

Entire original image (151KB)   

Unlike the 7100 and the 8100, it did not ship with an HPV card; instead, it always came with the PDS terminator. Don't lose it. No video card shipped as standard with the 9150.

Entire original image (127.4KB)   

No doubt about which model the board is for, but there is a lot of empty space on it. On the 1994 9150/120 Zephyr prototype, an ornate picture of a train and rollercoaster appears here, but production systems are blank.

Entire original image (142.7KB)   

Booted up into MacOS.

Entire original image (80.6KB)   


MkLinux on the WGS 9150

MkLinux was Apple's supported flavour of Linux, developed by OSF-RI in 1996 and sponsored by Cupertino. It is unusual in that it has a Red Hat-based userland but the underlying operating system is based on the Mach microkernel, which is atypical for Linux, but appropriate for Apple as both NeXTSTEP and OS X are based on it. (So is Tenon MachTen.) OSF planned to use MkLinux as a launching pad to get Mach on more platforms, and their involvement strongly hints that pieces of the abandoned Power Mac port of OSF/1, which is also based on Mach, were recycled for MkLinux. Eventually Apple transferred control to the community MkLinux Developers Association in 1998 to put internal resources towards OS X, which also used pieces of MkLinux to get it bootstrapped on Power Macs; it became superfluous as other Linux distributions started to support the same hardware with more conventional system designs, and the last preview release was R2 in 2002. The 9150 is specifically supported by MkLinux.

Entire original image (78.8KB)   

Linux Mac users will immediately recognize some form of this dialogue box, an extension used to select the bootloader at startup. Later OpenFirmware-based Power Macs can boot alternative kernels directly, and MkLinux supported machines all the way up to the early G4 series, but earlier beige Macs can only do so with difficulty and NuBus Macs can't at all, hence this stub. Unlike the booter used for, say, NetBSD/mac68k, it is an extension and not an application, so it can take over before any other extensions are loaded.

Entire original image (58.1KB)   

Startup messages. The COLOR message is particularly noteworthy -- if you boot early versions of Mac OS X through at least Jaguar on a colour framebuffer, this same message with the same colours appears on a verbose boot and is almost certainly descended from the same block of code. It disappeared by around 10.4 since every supported Mac used a colour framebuffer by then. Not all hardware works with MkLinux, by the way -- mine consistently kernel-panicked with an 8*24 (670) NuBus video card, even in 8-bit colour, so I had to use the built-in Ariel video.

Entire original image (89.8KB)   

The messages then end and the main startup begins on a blue screen. For some reason MkLinux uses the icky PC VGA font -- I would have preferred something a little more Mac-like.

Entire original image (184.3KB)   

Login prompt.

Entire original image (80.4KB)   

X does work on the 9150 without any special configuration, which is impressive (using the startx command). However, by default, it starts GNOME using Sawfish as the window manager, which is dreadfully slow; eventually I got sick of it and killed the X session to look at something else (Control-Option-Reset). AfterStep works pretty well, considering. You can try that by chmodding -x gdm and then putting exec /usr/X11R6/bin/afterstep in your ~/.xinitrc. On an ADB single-button mouse, press Option-2 and Option-3 for middle and right mouse clicks (the mouse button itself is a left click), which is directly emulated by the X driver. Mozilla 1.2.1 is included, but it's glacially slow, and Classilla is much more current. Plus, with the built-in Ariel video, you only have the choice of 832x624x8 (and get terrible colour clash) or 640x460x16 (and get no room to do anything). It's easiest to set the video mode in Mac OS first, which will then carry over to MkLinux.

Overall using X on MkLinux is not a lot of fun on the unexpanded 9150, but admittedly it would be running MkLinux in a server configuration anyway, not as a workstation.

Entire original image (88.5KB)   


generated by Floodgap Super Image Browser 1.788.0