Thursday, June 2, 2011

IBM RS/6000

IBM RS/6000's (RISC System 6000) are strange beasts (compared to more approachable machines from the same era such as anything made by Sun, Digital, HP, or SGI). But actually they're just different, not hostile.

Identifying

  • Every 'box' (computer, monitor, external tape drive etc.) has a designation of the form type-model (e.g., 7012-250). The tuple uniquely identifies the machine (it may still have differing CPU options). There might be an additional specifier (e.g., 7043-150 43P) which is just marketing and does not indicate a difference. Model specifications became more complicated with later models (after RS/6000 was renamed pSeries), but also more consistent and less confusing.
  • The type indicates the general class of machine all of which share the general case design (the case is not necessarily identical). The model designations are also mostly unique (there is only one 370, namely the 7012-370), with higher model numbers generally being higher-end models. There are some annoying exceptions, though. There are two kinds of 43P's: the 7248-43P, and the 7043-xxx, which IBM confusignly also refers to as 43P. They have the same case design, but are very different otherwise. Sometimes differing models just indicate the options the machine came with (e.g., 7006-41T vs 7006-41W which are identical except that one was shipped with a monitor). Sometimes, an additional letter is added to indicate a CPU upgrade: 320H is a 320 with a 25MHz CPU instead of the original 20MHz. Confusingly, this is often abreviated as 32H.
  • Every part has a part number that uniquely identifies the part. It may also have a FRU (field replacable unit) number. They both have the form NNXNNNN (N=number, X=letter), with no apparent system otherwise. Similar parts sometimes have similar part or FRU numbers. Some parts have no FRU number (they are not meant to be field replacable). The number ranges very rarely but annoyingly overlap, so it is important to note whether it's a part number or a FRU number.
  • Identical part numbers theoretically indicate that it's the same part.
  • Identical FRU numbers guarantee that the parts are compatible.
  • There might be several different parts associated with the same FRU ("it's not the same, but it will work").
  • A part may have several FRU numbers, and sometimes, not all of them are written on it. This is because some parts (e.g. memory carrier boards) can be used in many different machines, some of which might be unrelated.
  • All upgradable parts have an option number of the form #NNNN, which maps to one or several FRU numbers. They do not seem to be model specific (though they are only mentioned in the model specific documentation, but I have no example of an incosistency). The option number is used for ordering upgrades.
  • Optional adapters also have an "adapter type label" of the form N-X (X being a number or letter). The number before the dash is the general type, e.g., video (1), network (2), multiport async serial (3), disk (4), the whole label identifies the exact type of card, e.g., Ethernet High-Performance LAN Adapte (2-1), Token-Ring High-Performance Network Adapter (2-2), Thick/Thin Ethernet Riser (2-8), or FDDI-Fiber Single Ring Adapter (2-R).

Classification

Grey boxThe original RS/6000's were "IBM-grey". They came in different sizes (desktop, deskside, rackmount), had a POWER or POWER2, PowerPC-601, or RS64 processor, Microchannel slots (some later models had ISA/PCI slots instead). Except for marketing, there was no artificial distinction between workstations and servers
Black boxWith the advent of the PowerPC architecture, IBM switched to black cases. They have a processor from a PowerPC-604 or POWER3 upwards
SP"Scalable POWERparallel", IBM's first attempt at parallel computing. In principle they are RS/6000's, but SP "Nodes" can only be used with "Frames" and a HMC (Hardware Management Console).
OEMBull was selling IBM RS/6000's under their own name (and also the other way around, the first SMP RS/6000's were designed by Bull for IBM). These OEM boxes sometimes had very fancy casing not resembling IBM machines at all
PRePIn an effort to make the PowerPC architecture more mainstream, the PREP (PowerPC Reference Platform) was designed basically as a PC with a PowerPC processor. Some gray-box RS/6000 are PReP machines, and there have been PReP machines by Bull (Estrela), Motorola (PowerStack), and Apple (ANS, "Apple network Server", based on a modified PowerMac 9500). They were supposed to run several operating systems, including AIX (usually requiring a special version), OS/2 (never left beta), Solaris (2.5.1 only), Windows NT (up to 4.0), Taligent (never finished). Some later revision of the spec mandated the presence of OpenFirmware
CHRPAs everybody seemed rather unhappy with PReP (especially Apple), a new effort to standardize was started, and resulted in CHRP (Common Hardware Reference Platform). To please Apple, it was more similar to the PowerMac, but similarly unsuccessful as PReP. CHRP and PReP are not exclusive (a machine can be both).

Processor generations

POWER199020-62.5MHz"Performance Optimized With Enhanced RISC". It came in 3 varieties: the original 11-chip implementation (RIOS-1), a slightly reduced 8-chip (RIOS.9), and the (even cheaper and more reduced) "RISC Single Chip" (POWER RSC) variety
POWER2199355-160MHzA much enhanced (faster) implementation with a few added instructions. It was still a multi-chip design (6 or 8-chips), but later replaced by the POWER2SC (POWER2 Super Chip)
PowerPC-6xx199350-400MHzThe PowerPC architecture was designed in a joint effort between Apple, IBM, and Motorola ("AIM Alliance"). It is based on the POWER architecture, but with some instructions added, and some removed. It also added a 64bit instruction set and support for SMP. PowerPC-601 was a hybrid and implement both the PowerPC and the complete POWER instruction set
POWER31998200-450MHzBased on the unsuccessful PowerPC-620 (the first 64bit PowerPC, never used in a RS/6000), and implemented all optional instructions (so it is a superset of POWER2 and PowerPC). The Power3-II was an improved Power3 running at up to 450MHz
RS641997125-750MHzA variety of the "Amazon" AS/400 RISC processor, with a sufficiently complete PowerPC instruction set to be used with RS/6000's. The RS64-II (1998) was one of the first processors to implement multithreading, the RS64-III (1999) and RS64-IV (2000) ran at up to 750MHz
POWER420011.1-1.9GHzUnified the POWER3 and the RS64 and implemented the PowerPC 2.0 ISA
POWER520041.5-2.3GHzA dualcore design with SMT (simultaneous multithreading)
POWER620073.5-5GHzSuccessor of POWER5
POWER720102.4-4.25GHzUp to 8 cores, and reducing the GHz for the first time

Working with RS/6000's

  • The keyswitch present in early RS/6000's has 3 positions: normal, locked, and maintenance. Normal is what you'd expect, locked disables the keyboard and reset switch, and maintenance unlocks the case, and uses a different bootlist which usually allows booting from floppy, cdrom, tape, or network. The OS can check the state of the keyswitch and will boot into maintenance mode when it is set to maintenance. Some newer models have a "software-keyswitch" (which can be set to maintenance in the firmware).
  • Be patient, and check what the LED display says. Unlike most other Unix machines, early RS/6000 communicate with you only through a 3-digit LED display, until some kind of operating system has taken over. I.e., unless the machine has successfully booted an operating system, nothing is displayed neither on the display nor on the serial port. This is aggravated by the fact that they take comparatively long to boot. The display is actually very informative and tells you exactly in which state the machine is, and what it is currently trying to do. A list of codes can be found here. The LED turns blank when the operating system has been loaded successfully.
  • The HMC (Hardware management console) is used to control (power on/off, install os, access console) SP-class machines. It can also be used to control newer machines (but it is not strictly required as these also have a normal console). It is usually connected by a serial line.
  • On the serial port, AIX requires CTS and DCD (the installer does not). When the installation goes well, but after reboot the machine hangs at "Completed NFS services" (AIX 4 and 5) or "Multi-user initialization completed" (AIX 3), this is most likely your problem. Easiest is to wire CTS and DCD to DTR (on DB25: wire pins 4, 6, and 8 together).
  • RS/6000 are for the most part not at all picky about devices (they take most standard SCSI CDROM drives, harddisks etc.), but they require a special IBM keyboard. A random sampling of (non-rs6k) IBM branded keyboards showed that about half of them work. None of the non IBM branded keyboards I tried worked. The key point seems to be that the keyboard must contain a speaker and understand some additional initialization. There's nothing special about the mouse, any PS/2 mouse will do.
  • The installer does not try very hard to recover from errors, it will just print a an easily overlooked message and go on. Carefully check the log for errors and start over if there is anything unusual.
  • If the machine seems flaky, check the power supply voltage levels (especially that of external CD-ROM drives). Run diagnostics (from within AIX, any AIX CD, the special diagnostics CD, or diagnostic diskettes), they are very exhaustive. If this doesn't help, try the "Kreidler-Method" (named after the German Kreidler mopeds which often magically work again after being taken apart and put together again without really changing anything).

Links

No comments:

Post a Comment