Your answer to this question might be based on financial limits or it might be based on performance requirements. Keep in mind that any SP or clustered enterprise server system is scalable which means that you can add more nodes later. Your answers to the prior questions should have helped you determine the type of work for which you will be using the system. For example, if you determined that you want to use SP system partitions, it can affect the number of nodes you require. Since the system is scalable, you can select fewer nodes now and add more later or select more now and scale down later.
To decide which and how many nodes you need, consider the overall capacity of each node keeping in mind your function, performance, network, communication, and data transmission requirements, together with the software that you have selected to satisfy your needs, and which nodes can handle those combined demands. |There are the SP rack-mounted nodes (thin, wide, and high). |There are other processor nodes which can function like SP rack-mounted nodes, |but are servers that are not in an SP frame. There is the SP Switch |Router, an extension node to handle high data transmission demands in an SP |system that uses the SP Switch.
|Several worksheets originally for information about one SP frame at |a time, with rows that relate to the nodes in slots within that frame, have |been adjusted to accommodate servers as well. |The p690 server has characteristics similar to SP frames. Use |one copy of such worksheets for each p690 server with multiple LPARs. |Use a row for each LPAR node.
|For servers in a one-to-one frame and node relationship, including |unpartitioned p690 servers, use one copy of each worksheet to document up to |16 servers. Ignore the frame and switch number header fields and use a |row for each server. Use the fixed row number in the worksheet as a |sequence number and add the frame number you assign to each |server.
You can choose from the following |SP rack-mounted nodes:
The 375 MHz POWER3 SMP thin node can have two or four 64-bit 375 MHz processors, 256 MB to 8 GB of memory, two 32-bit PCI slots, a slot for the SP Switch MX2 adapter, integrated Ethernet, 4.5 GB to 18.2 GB of mirrored internal DASD, integrated Ultra SCSI, and an external RS-232 connection with active heartbeat used only by the HACMP application. This node is available as a single node. It does not have to be used in a pair. When it is a single node in a drawer, it must be in the odd-numbered slot.
|Thin nodes are supported in a short or tall SP frame in switchless, |SP Switch2, or SP Switch configurations. In a tall frame they can |connect to the SP Switch2 or the SP Switch. In a short frame they can |connect to the SP Switch-8.
Wide nodes greatly expand the I/O and network server functions of the SP system. Wide nodes occupy two slots in a frame and have more attachment options than thin nodes to increase DASD and network connectivity.
The 375 MHz POWER3 SMP wide node has two or four 64-bit 630+ floating point processors, 256 MB to 8 GB of memory, ten PCI slots (two 32-bit and eight 64-bit), a slot for the SP Switch MX2 adapter, integrated Ethernet, and 4.5 GB to 54.6 GB of mirrored internal DASD, integrated Ultra SCSI, and an external RS-232 connection with active heartbeat used only by the HACMP application.
|Wide nodes are supported in a short or tall SP frame in switchless, |SP Switch2, or SP Switch configurations. In a tall frame they can |connect to the SP Switch2 or the SP Switch. In a short frame they can |connect to the SP Switch-8.
A high node occupies two full drawers of an SP frame.
The 375 MHz POWER3 SMP high node can have pairs of 630FP (floating point) processors from four to sixteen, 1 to 32 GB of memory, five PCI slots (one 32-bit and four 64-bit), a slot for the SP Switch MX2 adapter, integrated 10/100 Ethernet, 9.1 to 36.4 GB of mirrored internal DASD, integrated Ultra SCSI, and an external RS-232 connection with active heartbeat used only by the HACMP application. It can have up to six SP Expansion I/O Units connected directly by cable for increased internal DASD and connectivity to external DASD and networks. On the other hand, you can choose to have no internal DASD and use only external DASD, even for booting.
|The 375 MHz POWER3 SMP High nodes are supported only in a tall SP |frame in switchless, SP Switch2, or SP Switch configurations.
|You can order certain IBM servers for running PSSP |3.4. Each one can be within an SP system, in which case it is |called an SP-attached server, or it can be part of a system of |clustered enterprise servers. These terms merely signify the |system configuration in which a node participates when it runs the PSSP |software. Except where otherwise noted, statements in this book |about function on SP nodes apply also to a node that is an SP-attached server |or in a clustered configuration. Functionally they are all simply nodes |in the system.
|A system is technically an SP system if there is at least one SP |frame with nodes in it. A clustered enterprise server is independent of |the SP frame. A clustered enterprise server system is comprised of a |control workstation running the PSSP software and connected to each of up to |32 servers also running the PSSP software. Such a system functions like |an SP system, but it does not have SP rack-mounted nodes. If your |system has at least one SP frame for an SP switch, you must understand and |follow the support and configuration conditions associated with that switch, |regardless of whether your system is an SP or a cluster of servers. |There are some limitations resulting from the hardware differences. |Except where otherwise noted, discussions in this book about function and |networks apply to a system of clustered enterprise servers as well as to an SP |system. Some layout and connection rules differ.
|The following characteristics significant to PSSP software and |configuration planning apply to any node configured as an SP-attached server |or a clustered enterprise server: |
|For example, if you have 16 servers with SAMI protocol, you can only have |16 with CSP protocol. You can have 8 servers with SAMI protocol and 24 |with CSP protocol. You can not have 24 SAMI servers and 8 CSP |servers.
|The following considerations are uniquely significant to planning |for running PSSP software in a cluster configuration: |
|Each of these servers have PCI-based 64-bit symmetric |multiprocessors of varied capacities delivering performance, scalability, and |reliability for today's critical e-business applications. Each |supports concurrent 32-bit and 64-bit application processing. You can |use any of the following as SP-attached servers or in a clustered enterprise |server system configuration: |
|The p690 is an 8- to 32-way SMP RS/6000 server. It can be configured |with 8, 16, or 24 1.0 GHz processors or with 16, 24, or 32 1.3 |GHz processors. These physical resources can be configured into up to |16 logical partitions (LPAR). The p690 is represented by PSSP as a |single frame. If it is not partitioned, it is represented as a single |node in that frame. If the physical resources of the p690 are |partitioned, each LPAR is a separate node in that frame and running its own |AIX 5L 5.1 image with the PSSP software. Every LPAR in the |server is recognized as a node - no provision is made in PSSP to ignore |any LPAR.
|Up to sixteen p690 servers, with a maximum of 48 total LPARs, can |participate as clustered enterprise servers or as SP-attached servers in any |one system. An unpartitioned p690 has one LPAR and is seen by PSSP as |one node. A partitioned p690 is seen by PSSP as one frame with as many |nodes as there are LPARs. The number of these servers counts toward the |total number of servers in one system which is not to exceed 32 and the total |number of node frames which is not to exceed 128. The total number of |LPARs count toward the maximum number of nodes in one system, which in a |standard SP system is 128. Additional constraints apply depending on |the switch configuration: |
|Connectivity from the SP control workstation to the p690 server is |through the Hardware Management Console (HMC) by a network connection to the |SP Ethernet admin LAN. The HMC is a processor that runs only the HMC |software for installation and service support of the p690. It is |connected through a serial connection to the Common Service Processor (CSP) |that is integrated in the p690 server. The HMC represents and manages |the p690 through the Common Information Model (CIM) and provides hardware |control capability and the function to logically partition the physical |resources in the server. For stand-alone p690 servers, the HMC is |optional if the LPAR function is not needed. For p690 servers that are |attached to the SP or are part of a system of clustered enterprise servers, |the HMC is required whether or not the server is configured for LPAR. |It is necessary for a single point of control from the control |workstation. One HMC can be used for multiple p690 servers. Each |p690 server can have redundant (two) HMCs.
|These servers use the HMC hardware protocol. No RS-232 cable |connecting to the control workstation is necessary.
|Since each LPAR functions as an SP node, PSSP requires that each LPAR be |configured with an Ethernet adapter connected to the SP Ethernet admin |LAN. Previously, PSSP required that the SP LAN connection be made |explicitly through the en0 adapter from the node. For the SP LAN |connection from this node, you can specify any Ethernet adapter that is |currently supported for connection to the SP LAN. The adapter can be |installed at any logical location on the device tree. You can specify |an adapter to be configured by its physical location and the PSSP software |will map the physical location to the logical device name when using AIX |commands to define and configure the adapter on the node.
|Each LPAR that is to use a switch needs to have a suitable adapter. |If you plan to use the SP Switch2 with two switch planes, each LPAR needs two |adapters, css0 to connect to the first switch plane and |css1 to connect to the second switch plane. If you plan to |use the SP Switch, you need to specify the switch port connections by using a |PSSP configuration file. The switch primary and backup nodes can be |assigned to LPARs on p690 servers. However, do not assign both the |primary and backup nodes in the same physical server if possible. Since |these LPARs share some common level of hardware and power controls, the entire |switch would not be available if that server is not available for any |reason.
|PSSP does not provide an interface to logically partition the p690 |server. You need to use the WebSM facility provided on the HMC |console. However, you can display the HMC graphical user interface on |the control workstation monitor. An interface to launch a remote WebSM |session to the HMC console is provided by SP Perspectives.
|An option is available through the HMC to logically partition the |p690 server such that the partitions contain processors, memory, and I/O that |are in physical proximity. Using this option results in more |predictable computing performance for these LPARs. It might be an |important consideration when the LPARs are used as compute nodes for large |parallel applications or in systems where predictable node performance is |required. See the book Hardware |Management Console Operations Guide for additional information on |configuring the p690 LPARs. |
|These can participate in a cluster of a total of 32 servers or up to |32 SP-attached servers in one system. Each is represented as a single |node in a single frame. It uses the Common Service Processor (CSP) to |provide hardware control and monitoring similar to the SP frame and node |supervisor. These servers use the CSP hardware protocol with one RS-232 |cable to the control workstation. The following are characteristics of |these models: |
|The M80 is a 19 inch rack-mounted node. |It can participate in a cluster of a total of 32 servers or up to 32 |SP-attached servers in one system. It is represented as a single node in a single frame. It |uses the Common Service Processor (CSP) to provide hardware control and |monitoring similar to the SP frame and node supervisor. The M80 is a |midrange, up to 8-way, 64-bit SMP server. A fully configured system |consists of one processor drawer and four I/O drawers for a capacity of 56 PCI |slots. |These servers use the CSP hardware protocol with one RS-232 cable to |the control workstation. |
|The pSeries 680 is a 24 inch rack-mounted node. |Up to 16 can participate in a cluster of a total of 32 servers or as |SP-attached servers in one system. Each is represented as a single node in a single frame. It |is an up to 24-way, 64-bit SMP server with state-of-the-art copper |silicon-on-insulator technology. A fully configured system has 24 600 |MHz RS64 IV SMPs, 96GB of memory, 56 PCI adapter slots, 48 hot-swappable disk |bays, 8 media bays, and 873.6GB of internal disk. |This server uses the SAMI hardware protocol with two RS-232 cables |to the control workstation. |
|Table 9 summarizes the nodes that you can currently order
|from IBM. They are all supported with the SP Switch2 or the SP
|Switch. The nodes that do not have to be in a drawer of an SP frame can
|be attached to an SP system or can be in a cluster system configuration where
|the control workstation and all nodes are running the PSSP software.
|See the book IBM RS/6000 SP: Planning Volume 1, Hardware and
|Physical Environment for hardware requirements.
|
|Table 9. SP and other nodes currently available from IBM with PSSP 3.4.
Node (min nodes to SP drawer) | Speed | Processing | Min to max memory | Min to max internal disk space | Max switch planes | PSSP level at announce |
---|---|---|---|---|---|---|
p690 1-16 LPARs | 1.0 or 1.3 GHz | 64-bit SMP w AIX 5L 5.1 | 8 to 256GB | 1 to 6 RIO | 2 | 3.4 |
p660 6M1 | 668 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 1 to 64GB | 4 to 8 RIO | 2 | 3.4 |
p660 6H1 | 450, 600, or 668 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 512MB to 32GB | 0 to 36.4GB | 2 | 3.2 |
p660 6H0 | 450 or 600 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 512MB to 32GB | 0 to 36.4GB | 2 | 3.2 |
p680 | 450 or 600 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 2 to 96GB | 9.1 to 873.6GB | 2 | 3.2 |
M80 | 500 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 1 to 32GB | 0 to 36.4GB | 2 | 3.2 |
POWER3 High (1-2) | 375 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 1 to 64GB | 0 to 946.4GB | 2 | 3.1.1 |
POWER3 Wide (1-1) | 375 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 256MB to 16GB | 0 to 109.2GB | 1 | 3.1.1 |
POWER3 Thin (1-1/2) | 375 MHz | 64-bit SMP w AIX 5L 5.1 or AIX 4.3.3 | 256MB to 16GB | 0 to 36.4GB | 1 | 3.1.1 |
|
|You might already have SP rack-mounted nodes or other servers that |are still supported and you can migrate to PSSP 3.4 with AIX 5L |5.1 or AIX 4.3.3. You can still use any of the |following nodes: |
|With PSSP 3.4, all the supported nodes except the S70 and S7A |can be connected to the SP Switch2. When all nodes are running the PSSP |3.4 software, the S70 and S7A can be in a system that has the SP |Switch2 but not connected to the switch. All the supported nodes can |connect to the SP Switch.
|Table 10 summarizes the nodes you might already have that
|can run PSSP 3.4 with AIX 5L 5.1 or AIX
|4.3.3. Nodes that do not have to be in a drawer of an SP
|frame might be attached to an SP system or might be in a cluster configuration
|where the control workstation and all nodes are running the PSSP
|software. If you plan to migrate an earlier level of PSSP to PSSP
|3.4 on nodes you already have or you plan to run mixed levels of PSSP,
|be sure to carefully read Chapter 11, Planning for migration.
|
|Table 10. Nodes you might already have that can run PSSP 3.4
Node (min nodes to SP drawer) | Speed | Processing at announce | Min to max memory | Min to max internal disk space | Max switch planes | PSSP level at announce |
---|---|---|---|---|---|---|
H80 | 450, 500, 600, or 668 MHz | 64-bit SMP w AIX 4.3.3 | 512MB to 32GB | 0 to 36.4GB | 2 | 3.2 |
S80 | 450 MHz | 64-bit SMP w AIX 4.3.3 | 512MB to 64GB | 4.5GB to 218GB | 2 | 3.1.1 |
POWER3 High (1-2) | 222 MHz | 64-bit SMP w AIX 4.3.3 | 1 to 16GB | 9.1 to 18.2GB mirrored | 2 | 3.1.1 |
POWER3 Wide (1-1) | 200 MHz | 64-bit SMP w AIX 4.3.3 | 256MB to 4GB | 4.5 to 36.4GB mirrored | 1 | 3.1 |
POWER3 Thin (1-1/2) | 200 MHz | 64-bit SMP w AIX 4.3.3 | 256MB to 4GB | 4.5GB to 18.2GB mirrored | 1 | 3.1 |
S7A | 262 MHz | 64-bit SMP w AIX 4.3.3 | 512MB to 32GB | 4.5GB to 218GB | 1 ¬sps2 | 3.1 |
S70 | 125 MHz | 64-bit SMP w AIX 4.3.3 | 512MB to 16GB | 4.5GB to 218GB | 1 ¬sps2 | 3.1 |
Wide (1-1/2) | 332 MHz | 32-bit SMP w AIX 4.2.1 or 4.3.2 | 256MB to 3GB | 4.5GB to 36.4GB | 1 | 2.4 |
Thin (1-1/2) | 332 MHz | 32-bit SMP w AIX 4.2.1 or 4.3.2 | 256MB to 3GB | 4.5GB to 18.2GB | 1 | 2.4 |
|
|Configuring a system using extension nodes requires special planning with |respect to processor nodes. The only unit currently in this category is |a dependent node, the SP Switch Router. Extension nodes are only |supported with the SP Switch, not with the SP Switch2.
|Regardless of which processor node types you use, if you plan to use the SP |Switch and order an SP Switch Router, you have to reserve one node slot for |each connection from the SP Switch Router to the SP Switch. This is |necessary to have a switch port available for each SP Switch Router |Adapter.
|Each SP Switch Router Adapter in an SP Switch Router must be connected to a |valid switch port on the SP Switch. To accommodate that requirement, |each dependent node logically occupies a slot in an SP frame and physically |occupies the switch port corresponding to that slot. A processor node |must not be assigned to the same slot, although a wide or high node can |overlay the slot. For a discussion of valid extension node slots read "Chapter 3, Defining the configuration that fits your needs" beginning with "Considering the SP Switch Router".
There are four SP frame models for the SP system which you can populate
with optional nodes and switches to create the SP system configuration of your
choice. Your layout can range from a single-frame starter system to a
highly parallel, large-scale system. |The frame models for SP nodes currently available from IBM are
|listed in Table 11. Other frames with no nodes are available
but not discussed here.
|
|Table 11. The basic SP node frames
Frame Model | Description |
---|---|
500 | Short base frame, power supply, additional equipment:
|
550 | Tall base frame, power supply, additional equipment:
|
1500 | Short expansion frame, same support as short base frame but has no prerequisite of a node |
1550 | Tall expansion frame, same support as tall base frame but has no prerequisite of a node. |
|
Now it's time to take all the information you have thought about and start to lay out your system requirements on detailed worksheets. These worksheets are an invaluable tool for helping you plan your configuration and installation in detail. If you have not done it already, make copies of the worksheets in Appendix C, SP system planning worksheets. The worksheets in this chapter have been filled out for a hypothetical customer, the ABC Corporation. The major system hardware components selected for the ABC Corporation are in Table 12.
|Table 12. Major system hardware components
Major system hardware components - Worksheet 4 | ||||||
---|---|---|---|---|---|---|
Company name: ABC Corporation | Date: November 20, 2001 | |||||
Customer number: 999999 |
| |||||
Customer contact: Jim Smith | Phone: 1-800-555-5678 | |||||
IBM contact: Susann Burns | Phone: 1-800-555-6789 | |||||
Complete the following by entering quantities to order: | ||||||
Frames | Nodes | Nodes | ||||
p690: | number LPARs: | |||||
550 (tall): 1 | 375 MHz High: 1 | p660 6M1: | ||||
1550 (tall): 1 | 375 MHz Wide: 2 | p660 6H1: | ||||
500 (short): | 375 MHz Thin: 6 | p660 6H0: | ||||
1500 (short): | M80: | p680: | ||||
| Switch subsystem components | |||||
| SP Switch2: 2 | SP Switch2 Adapter: 18 | SP Switch2 PCI-X Attachment Adapter: | |||
SP Switch2 MX2 Adapter: | SP Switch2 PCI Attachment Adapter: | |||||
| SP Switch 16-port: | SP Switch MX Adapter: | SP Switch MX2 Adapter: | |||
| SP Switch 8-port: | SP Switch Adapter: | RS/6000 SP System Attachment Adapter: | |||
| Extension node components | |||||
| SP Switch Router: | SP Switch Router Adapter: | ||||
| SP Expansion I/O Units: 6 | |||||
External storage units: | Type | Quantity | ||||
7133 | 32 | |||||
| ||||||
| ||||||
| ||||||
| Network media cards: | Type | Quantity | |||
| ||||||
| ||||||
| ||||||
| ||||||
Fill in the remaining fields after you place your order. | ||||||
| SP model number: 306 | SP serial number: 510077730 | Purchase order number: | |||
| Control workstation: | |||||
| Peripherals: | |||||
| ||||||
| ||||||
| ||||||
|
|Complete your copy of Worksheet 4, "Major system hardware |components" in Table 60 with the heading information, the number of frames, the |number of each node type, the number of switches and adapters, and other |components you want. If you selected a external disks in Question 6: What are your disk storage requirements?, copy the information from that table to your copy of |Worksheet 4. When you place your order for an SP system fill in the SP |model number, SP serial number, and the purchase order number for easy |reference. The SP model and serial number fields do not apply to a |system of clustered enterprise servers.
These instructions explain one way to document your node layout. First draw a diagram of your system. Then add network information to that diagram. After that, write your network information into your copies of the worksheets. Fill in as many copies of Worksheet 5, Figure 58, as you need. Use one copy for each SP frame or equivalent. An example network is shown in Figure 5 and Figure 6.
To complete the node layout worksheets, do the following:
|If you are planning an SP Expansion I/O Unit, extension node, or |SP-attached server, make up an indicator for each of them and mark the slots |they use. An SP Expansion I/O Unit physically occupies a slot but |leaves the switch port free. An extension node or SP-attached server |only logically occupies the slot but uses the associated switch port in a |system with the SP Switch. You can use the same slot for both an SP |Expansion I/O Unit and either an extension node or an SP-attached |server.
Figure 5 shows a single frame with numbered slots (terms in parentheses are switch port numbers).
Figure 5. A node layout example
View figure.
For instruction on node and switch port numbering, see Understanding placement and numbering.
At this point, your layout might look something like that in Figure 6.
Figure 6. A node layout example with communications information
View figure.
You need to record the hardware configuration of your frames. At the same time you decide what types and how many nodes and other units you want, you also need to decide and keep track of how many processors, how much processor memory, and how much internal disk storage each processor node will have. Each of these values will affect the performance of your system, so choose carefully.
|After you decide on this information, fill in your copy of Worksheet |7, "Hardware configuration by frame" in |Table 61. You need multiple copies of this |worksheet depending on what is to comprise your system. Use one copy of |this worksheet for each SP frame or equivalent like the p690 |server.
|For SP frames with nodes, this worksheet is intended for information |about one frame at a time and the rows relate to the nodes and SP Expansion |I/O Units in slots within that frame. Complete column three with node |type for SP rack-mounted nodes. You might want to put the frame number |there for SP-attached servers. To be thorough, include any SP Expansion |I/O Units. Record the frame and slot of the associated node, as |demonstrated by the entries for slots 15 and 16 in Table 13. Leave blank any fields or entries that do not |apply.
For p690 servers, use one copy for each server, completing up to sixteen rows in each copy. Each row is a node based on an LPAR. For servers in a one-to-one node and frame relationship, including unpartitioned p690 servers, use one copy of this worksheet to document up to 16 servers, using a row for each server. Use two copies to document up to the maximum of 32 servers in one system. Ignore the frame and switch number header fields and use each row as a frame. You might want to put the model in column three for SP-attached servers or for servers in a clustered configuration.
The ABC Corporation planned an SP system and made the choices in
Table 13. Fields that do not apply to their system are left
blank.
|
|Table 13. ABC Corporations's choices for hardware configuration by frame
Hardware configuration by frame - Worksheet 7 | |||||
---|---|---|---|---|---|
Frame number: 1 | Hardware protocol: SP | Switch number: 1 | |||
p690 server name: | HMC hardware monitor user id: | ||||
HMC IP addresses: | |||||
Slot or LPAR or frame | Node or Expansion number | Node type or Associated frame/slot | Number processors, memory | Internal disk | Additional adapters |
1 | 1 | wide | 4, 8GB | 36.4GB mirrored | TokenRing(1), SSA(4), Ethernet(1), ESCON(1) |
2 | -- | ||||
3 | 3 | wide | 4, 8GB | 36.4GB mirrored | TokenRing(1), SSA(4), Ethernet(1) |
4 | -- | ||||
5 | 5 | thin | 4, 8GB | 18.2GB mirrored | TokenRing(1), SSA(2) |
6 | 6 | thin | 4, 8GB | 18.2GB mirrored | TokenRing(1), SSA(2) |
7 | 7 | thin | 2, 3GB | 18.2GB mirrored | TokenRing(1), Ultra SCSI(2) |
8 | 8 | thin | 2, 3GB | 18.2GB mirrored | |
9 | 9 | high | 16, 32GB | 18.2GB mirrored | TokenRing(1), SSA(8), Ethernet(4), FDDI(2), SCSI(4) |
10 | -- | ||||
11 | -- | ||||
12 | -- | ||||
13 | 13 | thin | 2, 3GB | 18.2GB mirrored | TokenRing(1), Ultra SCSI(2) |
14 | 14 | thin | 2, 3GB | 18.2GB mirrored | TokenRing(1), Ultra SCSI(2) |
15 | 15 | 1/9 |
|
| |
16 | 16 | 1/9 | |||
|
Each adapter in each node, workstation, and router has an IP address. Each of these addresses has a separate name associated with it. The SP system uses only IPv4 addresses. Some PSSP components tolerate IPv6 aliases for IPv4 network addresses but not with DCE, HACMP, HACWS, or an SP switch. For information about the SP system tolerating IPv6 aliases for IPv4 network addresses, see the appendix on the subject in the book PSSP: Administration Guide.
During installation and configuration, all addresses, including the router addresses, must be resolvable into names. Likewise, all names both long and short, must be resolvable into addresses. If your network administrator or support group provides name-to-address resolution through DNS, NIS, or some other means, they need to plan for the addition of all these names to their servers before the system arrives. You must specify these names during configuration to be set in the PSSP System Data Repository (SDR). Since AIX is case sensitive, the names must match exactly.
Independent of any of the network adapters, each |node has a host name. Usually the host name of a |node is the name given to one of the network adapters in the |node.
The host name in the worksheet is referring to the name given to that adapter. You need to select which of these adapter host names is to be the one given to the |node. Mark the column of the adapter that will be the host name. While completing these worksheets, keep the following criteria in mind:
Review your network topology and fill in your copies of worksheets 8 and 9 which start with Table 62. Be sure to make extra copies before you complete them. If you have additional network adapters planned for some or all of your nodes, you need to plan their network information also. See the book IBM RS/6000 SP: Planning Volume 1, Hardware and Physical Environment for information about required and optional adapters.
|For SP frames with nodes, these worksheets are intended for |information about one frame at a time and the rows relate to the nodes in |slots within that frame. For p690 servers, use one copy for each |server, up to two servers, completing up to eight rows in each copy. |Each row is a node based on an LPAR. For servers in a one-to-one node |and frame relationship, including unpartitioned p690 servers, use one copy of |this worksheet to document up to 16 servers, using a row for each |server. Use two copies to document up to the maximum of 32 servers in |one system. Ignore the frame and switch number header fields and use |each row as a frame. Use the fixed row number in the worksheet for |sequencing, and enter the frame number you assign to the server.
|For all nodes but those on a p690 server, you must use the adapter |name en0. For nodes on a p690 server you can use |en0, another name, or the physical location, like en2 or |U1.9-P1-I2/E1, of any Ethernet adapter that is supported in |the node for connection to the SP Ethernet admin LAN.
The ABC Corporation completed their network configuration
worksheets starting with the "SP Ethernet admin LAN configuration" in Table 14. They chose to plan only the additional token ring
and ESCON connections at this time. They also completed the worksheet
for additional node network adapters shown in Table 15.
|
|Table 14. ABC Corporation's SP Ethernet admin LAN
SP Ethernet admin LAN configuration - Worksheet 8 | ||||
---|---|---|---|---|
Company name: ABC Corporation | Date: November 20, 2001 | |||
Frame number: 1 | p690 server name: | |||
Associated node slot or frame | Admin LAN netmask: 255.255.255.192 (en0 for all but p690) | Default route | ||
Hostname | Adapter name or physical location | IP Address | ||
1 | spnode01 |
| 129.40.60.1 | 129.40.60.125 |
2 | -- |
|
|
|
3 | spnode03 |
| 129.40.60.3 | 129.40.60.125 |
4 | -- |
|
|
|
5 | spnode05 |
| 129.40.60.5 | 129.40.60.125 |
6 | spnode06 |
| 129.40.60.6 | 129.40.60.125 |
7 | spnode07 |
| 129.40.60.7 | 129.40.60.125 |
8 | spnode08 |
| 129.40.60.8 | 129.40.60.125 |
9 | spnode09 |
| 129.40.60.9 | 129.40.60.125 |
10 | -- |
|
|
|
11 | -- |
|
|
|
12 | -- |
|
|
|
13 | spnode13 |
| 129.40.60.13 | 129.40.60.125 |
14 | spnode14 |
| 129.40.60.14 | 129.40.60.125 |
15 | -- |
|
|
|
16 | -- |
|
|
|
Notes:
|
|Table 15. ABC Corporation's additional adapters node network
Additional adapters node network configuration - Worksheet 9 | ||||
---|---|---|---|---|
Company name: ABC Corporation | Date: November 20, 2001 | |||
Frame number: 1 | p690 server name: | |||
Token ring speed: 16 | ||||
Associated node slot or frame | Additional adapters netmask: 255.255.255.192 | Default route | ||
Adapter name or physical location | Hostname | IP address | ||
1 | tr0 | sptok01 | 129.40.61.1 | 129.40.60.125 |
2 | -- |
|
|
|
3 | tr0 | sptok03 | 129.40.61.3 | 129.40.60.125 |
4 | -- |
|
|
|
5 | tr0 | sptok05 | 129.40.61.5 | 129.40.60.125 |
6 | tr0 | sptok06 | 129.40.61.6 | 129.40.60.125 |
7 | tr0 | sptok07 | 129.40.61.7 | 129.40.60.125 |
8 |
|
|
|
|
9 | tr0 | sptok09 | 129.40.61.9 | 129.40.60.125 |
10 | -- |
|
|
|
11 | -- |
|
|
|
12 | -- |
|
|
|
13 | tr0 | sptok13 | 129.40.61.13 | 129.40.60.125 |
14 | tr0 | sptok14 | 129.40.61.14 | 129.40.60.125 |
15 | -- |
|
|
|
16 | -- |
|
|
|
Notes:
|
The advantage of an SP switch is that it has its own subnet. You need to plan this switch network whenever you plan to use any of the following:
Do you plan to enable ARP over the switch? If not, you need to derive the switch IP addresses from the address of the first node plus the switch port number.
|A two-plane SP Switch2 system has two sets of switches and two |adapters per node. The switch planes are disjoint - each is |cabled exactly like a single plane and communication across the pair of planes |is achieved via software striping. In a two-plane SP Switch2 |configuration, the first SP frame has a switch to be configured as plane 0 and |the first expansion frame has a second switch to be configured as plane |1. Every node that will be connected to the switch has two |adapters. The first switch adapter in the node is to be configured as |css0 and connected to plane 0. The second switch |adapter in the node is css1 and connected to plane |1.
|Although each node has access to both switch planes, there is no |physical connection between the planes. The switch subsystem provides |an aggregate or multi-link IP device abstraction for the SP Switch2 |networks. This pseudo-device driver enables you to address the two SP |Switch2 planes with one IP address. The purpose for this virtual-device |interface, the third IP address to be configured as ml0, is to |allow IP messages to be transmitted in a more economical manner called |striping. The striping technique provides a capability to transmit |consecutive IP data across two fully operational adapters. It takes |advantage of the combined bandwidth of both adapters. For example, when |an IP message is sent between nodes and both nodes have access to both |available switch networks, consecutive datagrams are sent in a pattern like |adapter0, adapter1, adapter0, |adapter1, .... Using ml0 can ensure |that a single failure in the SP Switch2 subsystem does not cause a complete |outage to a node or other subsystem that is dependent on a switch. If a |fault occurs between a node and one of the two switch planes, a transparent |failover condition occurs using the ml0 interface in order to |access the remaining functional switch plane. For example, if |adapter0 malfunctions the resulting data flow would be |adapter1, adapter1, adapter1, |adapter1, .... To use this feature, plan to use |the spaggip command or the SP Configuration Database Management |SMIT tool after you configure css0 and css1 for each |node that is to be attached to the SP Switch2.
|Make copies of the switch configuration worksheet in Table 64 and complete it for your system. Before you
|start, see Switch port numbering and IP address assignment for additional guidance. If the
|hypothetical ABC Corporation were to use the SP Switch2 with two
|switch planes, their completed worksheet might look like that in Table 16.
|
|Table 16. ABC Corporation's choices for the switch configuration worksheet
Switch configuration - Worksheet 10 | |||||||
---|---|---|---|---|---|---|---|
Frame number: 1 | Switch number: 1 | css0 netmask: 255.255.255.192 | css1 netmask: 255.255.255.192 | ml0 netmask: 255.255.255.192 | |||
Slot number | Switch port number | css0 hostname | css0 IP address | css1 hostname | css1 IP address | ml0 hostname | ml0 IP address |
1 |
| spsw01 | 129.40.62.1 | spsw101 | 129.40.63.1 | spml01 | 129.40.64.1 |
2 |
|
|
|
|
|
|
|
3 |
| spsw03 | 129.40.62.3 | spsw103 | 129.40.63.3 | spml03 | 129.40.64.3 |
4 |
|
|
|
|
|
|
|
5 |
| spsw05 | 129.40.62.5 | spsw105 | 129.40.63.5 | spml05 | 129.40.64.5 |
6 |
| spsw06 | 129.40.62.6 | spsw106 | 129.40.63.6 | spml06 | 129.40.64.6 |
7 |
| spsw07 | 129.40.62.7 | spsw107 | 129.40.63.7 | spml07 | 129.40.64.7 |
8 |
| spsw08 | 129.40.62.8 | spsw108 | 129.40.63.8 | spml08 | 129.40.64.8 |
9 |
| spsw08 | 129.40.62.9 | spsw109 | 129.40.63.9 | spml09 | 129.40.64.9 |
10 |
|
|
|
|
|
|
|
11 |
|
|
|
|
|
|
|
12 |
|
|
|
|
|
|
|
13 |
| spsw13 | 129.40.62.13 | spsw113 | 129.40.63.13 | spml013 | 129.40.64.13 |
14 |
| spsw14 | 129.40.62.14 | spsw114 | 129.40.63.14 | spml014 | 129.40.64.14 |
15 |
|
|
|
|
|
|
|
16 |
|
|
|
|
|
|
|
|
When you plan to use an SP-attached server, you need to fill in the switch worksheet to set a switch port number even in a switchless SP system. This is because of the limited hardware interface to SP-attached servers. The SP functions cannot always derive all the information it needs like it can for SP nodes. During the SP installation and configuration process of your frames and nodes you will be asked to supply that number along with other values you are preparing during this planning phase.
|Keep in mind that switch node numbers are used for nodes and |SP-attached servers in all types of SP and clustered enterprise server |systems, including those systems with no switch. It is important to |recognize that the algorithms for assigning switch node numbers to nodes and |SP-attached servers differ depending on the type, or lack of, switch in the |system. See Switch port numbering for a discussion on how switch node numbers are assigned for |each system and switch type.
|This knowledge becomes critical when you are adding an SP Switch to |a switchless system that has SP-attached servers. The algorithm for |assigning switch node numbers changes, and the switch node numbers that you |use for SP-attached servers in your switchless system might not be valid in |the same system with an SP Switch. When first assigning switch node |numbers to your SP-attached servers in a switchless SP system, consider if you |might ever add an SP Switch to the system in the future. If you might, |try to choose switch node numbers for the SP-attached servers that will be |valid in both the switchless and SP Switch environments.
|For example, if each existing SP frame were to have an SP Switch
|added to it, a reasonable number would be one that is available because a wide
|or high node is overlaying the associated slot. The worksheet for that
|node might be similar to that in Table 17.
|
|Table 17. Example of switch configuration worksheet for SP-attached server in a switchless SP system
Switch configuration - Worksheet 10 | |||||||
---|---|---|---|---|---|---|---|
Frame number: 4 | Switch number: | css0 netmask: | css1 netmask: | ml0 netmask: | |||
Slot number | Switch port number | css0 hostname | css0 IP address | css1 hostname | css1 IP address | ml0 hostname | ml0 IP address |
1 | 27 |
|
|
|
|
|
|
2 | -- |
|
|
|
|
|
|
· · · | |||||||
16 | -- |
|
|
|
|
|
|