• Selection of system components must be made in the context of total application requirements. Although the configuration of system components must be done in steps (for example, base packages, CPUs, memories, etc.), these steps cannot be done in isolation.

• The order in which requirements are assessed is also important, since one requirement may impact others. Before proceeding, it would be useful to assess the total application requirements in the following order:

• What level of availability is required?
• If no single points of failure are allowed, then the solution should be configured as a multi-system cluster.
• If access to specific devices must be assured, consider redundant adapters - RAID, N+1 power, redundant PCI drawers, and redundant consoles.
• If software redundancy is required, consider clusters and/or hardware partitioning. The choice of hardware partitioning will generate a need for multiple master PCI drawers, multiple consoles, and I/O adapters.

• If the ² CPU On-Line Add and Remove² feature is required, refer to document EK-GSHPG-RM for configuration and operational requirements.

• What level of hardware partitioning is required for optimal system management?

• What overall capacities are required in terms of processor performance, memory capacity, and disk storage?

• How should the system be configured to optimize performance?
• In most cases, optimum performance is achieved if the system resources (CPUs, memory, and I/O adapters) are balanced across the quad building blocks in the system.
• Memory should be configured according to application guidelines listed in Step 4.

• What are the near-term system expansion needs?

• How will system cabinets be physically arranged? This will determine if expansion cabinets are required and what cable lengths are required.

• Mandatory purchase: The system cannot function without this option or service - the option or service must be ordered with the system.

• Required to function: This option or service is needed to support a working system - the option or service must be ordered with the system or be available onsite.

• Recommended: System performance or function will be enhanced if this option or service is ordered.

• Base system with operating system license (either OpenVMS or Tru64 UNIX), which includes one 1224-MHz CPU module

• Minimum of one memory module

• Software media and documentation for first system onsite

• Installation and/or startup services

• System management console or device and software with equivalent functionality

• CarePaqä Priority Service Plan

• VIS Services

• AlphaServer GS160 base systems contain one CPU module. Additional SMP CPUs may be added, up to the limits shown in above table. SMP CPU options include an operating system SMP license.

• AlphaServer GS160 base systems can be configured with optional HP Capacity on Demand (CoD) CPUs for non-disruptive future capacity expansion. The CPUs will be field installed as part of the system installation. The total number of CPUs – base CPU, SMP CPUs, and CoD CPUs – must adhere to the limits shown in the above table. Refer to the HP Capacity on Demand Program described in the ² Upgrades² section.

• Memory options are engineered specifically for use with this series and include additional components, which are integral to the system architecture.

• Memory options consist of a series of base modules that contain one memory array. A second array (called ² upgrades² in the table) may be added to a base module in the factory or in the field.

• Configuring for capacity: The highest capacity is achieved when the 3X-MS8AA-DB/DU combination is used.
• Configuring for performance: Interleaved operations reduce the average latency and increase the memory throughput over non-interleaved operations. Each memory base module is capable of 4-way interleaving with one array (no upgrades added) or 8-way interleaving with two arrays (base module plus one upgrade). A QBB configured with eight arrays (four base modules plus four array upgrades) provides 32-way interleaving and has the maximum potential memory bandwidth. Refer to ² Memory Applications Examples² below to determine which applications gain the most benefit from this bandwidth.
• Memory modules should be configured in powers of 2: That is, 0, 1, 2, or 4 base modules in a QBB. Upgrades should also be installed in powers of 2: 0, 1, 2, or 4 base modules in a QBB.
• Although mixed-capacity memory modules may be configured, the highest bandwidth is achieved when a QBB is populated with eight identical arrays: four base modules, and four upgrades. The next-highest bandwidth would be four base modules (four arrays).
• If it is not possible to match the capacities of all the arrays, the next best choice is to configure pairs of identical base modules, or base module/upgrade combinations. For example, a configuration of two 2-GB base modules (3X-MS8AA-CB), each with a 1-GB upgrade (3X-MS8AA-BU), is a better choice than a configuration of three 2-GB modules (3X-MS8AA-CB).

The following table represents how 8 GB could be configured in a 4-CPU QBB in each of the four referenced applications. The numbers under each application represent how many of each memory option would be ordered.

• A single AlphaServer GS160 can be divided into logical hardware partitions, each running an instance of Tru64 UNIX or an instance of OpenVMS. Each partition is allocated its own dedicated ² shared-nothing² set of hardware resources: QBB(s), CPU module(s), memory module(s), and I/O.

• Multiple-QBB hard partitions within a GS server do not provide complete hardware failure isolation across hard partitions. Single hard partitioned QBBs within the server do provide hardware failure isolation.

• Each hardware partition is viewed as a unique node, from a system point-of-view, with its own instance of Tru64 UNIX or OpenVMS operating system and application software, independent system console, and error log.

• Hardware partitions are defined on QBB boundaries; each partition is an integer multiple of QBBs.

• Up to two hardware partitions are supported on GS160 Model 8 systems; up to four hardware partitions are supported on Model 16 systems.

• One system management console (3X-DS8BA-xx or 3X-DS8DA-xx) and one console hub (3X-DS8AA-AA) are recommended per system.

• Supported option rules apply for maximum configurations of each AlphaServer GS160 system partition. Care must be exercised to ensure that any planned reconfiguration of hardware partitions will not violate option support rules.

• One Alpha 21264 6/1224-MHz CPU module

• One 3X-MS8AA-BB/CB/DB memory module (1 GB, 2 GB, 4 GB)

• One 3X-KFWHA-AA system I/O module and one 3X-DWWPA-AA master PCI drawer. Depending upon the configuration, this may require the use of an 3X-H9A20-AD/AE/AF expansion cabinet.

• AlphaServer GS160 systems are normally configured according to standard module placement rules, and are shipped with one copy of the operating system installed at the factory (Tru64 UNIX or OpenVMS). However, systems with hardware partitions offer hardware and software configuration flexibility. Factory integration services (VIS) are recommended to enable custom module configuration and factory installation of multiple copies of the operating system on hardware- partitioned systems.

• Balance the resources in the system (or hardware partition) based upon the available backplane space and the proposed option populations:
• Sparsely configured systems, those using half or less than half of their available capacity for CPUs, memory, and PCI drawers, should be configured with the options concentrated in as few QBBs as possible. For example, a GS160 Model 16 with eight CPUs, eight memory modules, and four PCI drawers would usually be configured in the first two QBBs. The first two QBBs would be ² active² and the 3rd and 4th QBBs would be available for expansion.
• Densely populated systems, those using more than half of their available capacity for CPUs, memory, and PCI drawers, should be configured with the options spread out across all QBBs.

• Configure active QBBs symmetrically, each with CPUs, memory, and PCI drawers.

• Configure the I/O adapters so that each active QBB has direct access to the most frequently accessed data.

• Base systems include operating system license (Tru64 UNIX or OpenVMS) that licenses hardware partitions up to the physical limit of the base system package: two hardware partitions for Model 8 systems, four partitions for Model 16 systems.

• If a product is licensed for 200 concurrent users, these users can be split among the partitions, but cannot exceed 200 total users.

• If users have an enterprise capacity license for a product, that license can be loaded into the license databases on each of the hardware partitions.

• If the system requires both OpenVMS and Tru64 UNIX operating systems be licensed, one operating system license is included in the base system and the second is added as a line item. The second operating system license upgrade, which includes the license for only one CPU, would be added to the order using the following part numbers. Order appropriate media and documentation kits from Step 13.

• Only those SMP processors intended for use with the second operating system must be similarly licensed. Use the following license-only part numbers to add an SMP license for any CPUs intended for use with the second operating system:

• The order of licensing is not important, but the following examples are similarly constructed for clarity. The configuration starts with a Tru64 UNIX base system part number and the addition of OpenVMS licenses.

• Base system order would include a DA-160CE-Ax and 15 3X-KN8BA-AD SMP upgrade CPUs
• Add one QB-63PAQ-AG OpenVMS Software Upgrade and 15 QL-MT1A9-6R OpenVMS Alpha SMP licenses

• Base system order would include a DA-160CE-Ax and 15 3X-KN8BA-AD SMP upgrade CPUs
• Add one QB-63PAQ-AG OpenVMS software upgrade and seven QL-MT1A9-6R OpenVMS Alpha SMP licenses

• User and capacity-based licenses would be added for the second operating system environment as though it were a standalone system.

• Power supplies included with Model 8 and Model 16 systems can support all combinations of CPUs, memory, and I/O that can be configured within the system boxes.

• Additional 48V power regulators can be ordered to provide N+ 1 power redundancy.

• For Model 8 systems, order one power supply to achieve N+ 1 capability; for Model 16 systems, order two power supplies to achieve N+ 1 capability.

H7506-BA

• One additional PCI drawer (master or expansion)

• One additional PCI drawer (master or expansion) and one DS-SL13R-xx StorageWorksä shelf, or

• One or two StorageWorks DS-SL13R-xx shelves

• Mixed configurations of BA36R and DS-SL13R-xx shelves are supported

• System power cabinet provides space for up to two forward facing storage shelves.
• Up to two DS-SL13R-xx Ultra3 SCSI (LVD) shelves; each shelf supports a maximum of 14 Ultra3 disk drives

• Each single-bus Ultra3 shelf requires a 3X-KZPCA-AA Ultra2 (LVD) SCSI adapter or DS-KZPCC-xx RAID controller and a SCSI cable to connect controller to shelf

• Each split-bus Ultra3 shelf requires two 3X-KZPCA-AA Ultra2 (LVD) SCSI adapters, at least one dual-channel 3X-KZPEA-DB Ultra3 (LVD) SCSI adapter, or DS-KZPCC-xx RAID controllers and SCSI cables to connect controller to shelf

• Ultra3 shelves connected to 3X-KZPCA-AA adapters in the power cabinet require BN38C-02 2-meter cables; DS-KZPCC-xx RAID controllers require BN37A-02
  2-meter cables.

• Ultra3 shelves connected to 3X-KZPCA-AA adapters in an attached expander cabinet require BN38C-10 10-meter cables; DS-KZPCC-xx RAID controllers require BN37A-10 10-meter cables.

• Ultra3 shelves connected to 3X-KZPCA-AA adapters in a remote expander cabinet require 10 20-meter BN38C-xx cables, depending upon physical cabinet location; DS-KZPCC-xx RAID controllers require BN37A-xx cables.

• Ultra3 Universal drives are listed in a subsequent section

• Additional power supply provides N+ 1 power for 4314R Ultra3 (LVD) StorageWorks shelves; power supply uses a dedicated location in the shelf.

• Not required for 4354R shelves.

• Model 8 systems support up to four PCI drawers; Model 16 systems support up to eight PCI drawers. One PCI drawer included in Model 8 and Model 16 base systems.

• Model 8 and Model 16 power cabinets provide space for one additional PCI drawer if no more than one internal storage shelf is configured.

• Additional PCI drawers and storage shelves can be configured in
  3X-H9A20-AD/AE/AF I/O expansion cabinet, described in a subsequent section.

• All PCI drawers contain 14 PCI slots configured into four PCI buses; two of the buses have four slots each, the other two buses have three slots each.

• There are two types of PCI drawers: expansion drawers and master drawers. Base system configurations include one PCI master drawer with 12 configurable PCI slots.

• Expansion drawers contain 14 PCI slots and N+1 redundant power system; expansion drawers are used for most PCI expansion applications.

• Optional master drawers contain 13 configurable PCI slots, N+1 redundant power system, plus the console ports and storage devices required for use as a system console. (These devices are listed on page 2. Note that the Fast Ethernet adapter is not included in optional master PCI drawers.) Optional master drawers have two applications:

• As redundant console sub-systems
• As consoles for individual partitions in hardware partitioned systems

• PCI drawers are connected to a QBB utilizing a 3X-KFWHA-AA system I/O module that connects to the PCI drawer using two BN39B cables.

• PCI drawers are connected to a QBB utilizing a 3X-KFWHA-AA system I/O module that connects to the PCI drawer using two BN39B cables.

• Maximum one additional drawer in the system power cabinet; see ² External Expansion Cabinets² for more details.

• PCI drawers can be split between multiple QBBs as long as all QBBs are contained within the same hardware partition.

• PCI drawers mounted in a common H9A20 Expansion Cabinet can server multiple systems.

• Additional PCI drawers and storage shelves can be installed in optional 3X-H9A20-AD/AE/AF expansion cabinets. Up to four
  3X-H9A20-AD/AE/AF cabinets are supported.

• 3X-H9A20-AD/AE/AF I/O expansion cabinet can be configured to hold all disk BA36R StorageWorks shelves or
  DS-SL13x-xx Ultra3 StorageWorks shelves or combination of StorageWorks shelves and PCI drawers.
• If no PCI drawers are configured, cabinet supports up to eight BA36R or five DS-SL13R-xx StorageWorks shelves.
• If one PCI drawer is configured, cabinet supports up to five BA36R or four DS-SL13R-xx StorageWorks shelves.
• If two PCI drawers are configured, cabinet supports up to four BA36R or three DS-SL13R-xx StorageWorks shelves.
• If three PCI drawers are configured, cabinet supports up to two BA36R or two DS-SL13R-xx StorageWorks shelves.
• If four PCI drawers are configured, the cabinet supports one BA36R or DS-SL13R-xx StorageWorks shelf.

• BA36R and DS-SL13R-xx StorageWorks shelves can be combined in the same expansion cabinet.

• If large quantities of disks are required, the use of StorageWorks Storage Array cabinets and components is highly recommended.

• Systems installed in the US and Canada may use the 3X-H9A20-AD when 120V input power is required. In all other cases, the
  3X-H9A20-AF is preferred because of the ability to support dual AC input.

• 3X-H9A20-AD/AE/AF cabinets may be joined to GS160 systems. PCI drawers placed in these cabinets require 7-meter I/O cables.

• 3X-H9A20-AD/AE/AF cabinets may be placed up to 6 meters from the system cabinet. Multiple expander cabinets may be connected to one another or placed separately. Each group of free-standing H9A20 cabinets requires an end-panel trim kit (CK-H9A20-AB).

• PCI drawers placed in remote cabinets require 10-meter I/O cables.

• Tru64 UNIX supports a maximum of 64 total SCSI controllers per operating system instance (hardware partition). OpenVMS supports a maximum of 26 total SCSI controllers per operating system instance. Total SCSI controllers (all types) in the system must be within these limits regardless of the maximum per system I/O adapter limitations. Refer to the ² Supported Options List² for specific configuration rules.

• Each master PCI drawer contains embedded SCSI controllers (a FIS disk and a CD), which is included in the overall count of SCSI controllers configured in the system (or partition). Tru64 UNIX counts FIS disk and CDROM as an embedded SCSI device. OpenVMS counts the FIS disk only as an embedded SCSI device. Therefore, one (OpenVMS) or two (Tru64 UNIX) SCSI controllers per master PCI drawer must be included in the total count of SCSI devices in the system.

• Calculating the total number of SCSI controllers in the system (or partition) is done by adding all the devices in the system that the operating system categorizes as a SCSI device. Tru64 UNIX includes the following devices in this count: KZPBA-CA, KZPBA-CB, 3X-KZPBA-CC, 3X-KZPCA-AA,
  DS-KZPCC-CE, DS-KGPSA-CA., DS-KGPSA-DA Fibre Channel, and two embedded master PCI components per master PCI drawer. OpenVMS includes the following devices in this count: KZPBA-CA, KZPBA-CB, 3X-KZPBA-CC, 3X-KZPCA-AA, 3X-KZPEA-DB, DS-KZPCC-AC, and one embedded master PCI component per master PCI drawer.

• For cluster configurations, use Y cable (BN39A-0G).

• Manufacturing may substitute correct cable lengths depending on configuration.

NOTE: OpenVMS 7.2-2, or later, is required; Tru64 UNIX 5.1B PK4, or later is required, maximum cable length is 12 meters.

3X-BC56J-xx

257319-B21