Home > Articles > Operating Systems, Server > Solaris

📄 Contents

  1. Module 1: Hardware Setup
  2. Sun Cluster 3.0 Hardware Planning Considerations
  • Print
  • + Share This
Like this article? We recommend

Sun Cluster 3.0 Hardware Planning Considerations

Sun Cluster 3.0 Design Goal

Sun Cluster 3.0 hardware setup moves beyond Sun Cluster 2.2 hardware setup. By design, Sun Cluster 2.2 hardware setup provided a platform for availability and limited scalability. In contrast, the Sun Cluster 3.0 hardware setup platform (as configured during these lab exercises) creates a cluster environment capable of hosting HA applications that can be architected to provide near continuous availability and scalability, both horizontal and vertical.

Customer Requirements (Analysis)

CAUTION

For a given set of customer requirements, Sun Cluster 3.0 hardware applications are implemented across many layers, each layer having several aspects that need to be considered when proposing a solution. Only a comprehensive analysis of the customer's application requirements combined with a thorough knowledge of Sun Microsystems products and services can ensure that the cluster solution meets the customer's internal service level agreement (SLA). An incomplete analysis often leads to a poorly functioning cluster that adds little value to the customer's business and actually increases their cost of ownership. Dissatisfied customers are unlikely to purchase further clusters from Sun Microsystems. See ''References'' on page 19 for additional references.

Architectural Limitations

The Sun Cluster 3.0 hardware architecture is able to provide the highest levels of availability for hardware, operating system, and applications without compromising data integrity. The Sun Cluster 3.0 hardware environment (for example, hardware, operating environment, framework, API, and applications) can be customized to create HA applications.

Regardless of design, any software or hardware has physical or technical limitations. The Sun Cluster 3.0 hardware setup environment can be used in varying hardware and software configurations. However, before implementing the Sun Cluster 3.0 software, it is important to understand limitations of the product.

Configuring Clusters for HA: Primary Considerations

The primary configuration and planning considerations for HA applications and databases include identifying requirements for: software versions and features, boot environment, shared storage, and data services and their agents.

Designing a production cluster environment for a mission-critical environment is a complex task, involving the careful selection of optimum components amid a seemingly baffling array of options. We recommend that you work closely with a qualified consulting practice, such as Sun Professional Services, in making these selections. For example, determining the optimum number or mix of database instances or services per node, or ensuring that no potential agent conflicts exists and resolving any service level conflicts.

Different cluster topologies require carefully prescribed setup procedures in relation to the following cluster components:

  • Number of logical hosts per node (including their agents, agent interoperability, and service level requirements)

  • Type of Volume Manager

  • Disk striping and layout

  • File systems vs. Raw Device database storage

  • Performance (local storage vs. GFS considerations)

  • Network Infrastructure requirements and redundancy

  • Client Failover Strategy

  • Logical host Failover Method (manual vs. automatic)

  • Naming conventions (host ID, disk label, disk groups, meta sets, mount points, etc.)

  • Normal Operations Policies and Procedures

  • Backup and Recovery Procedures for the SunPlex components

    NOTE

    See ''References'' on page 19, for relevant Sun BluePrints Online articles, when considering each of these important topics.

Standard Sun Cluster Configurations

The Sun Cluster 3.0 hardware configuration implemented for these hands-on labs represents an entry-level platform, providing no SPOF for the cluster pair.

No SPOF implies that a single component failure within the cluster platform will not permanently disable operations. Instead, the implemented failover mechanisms must be configured to enable the surviving node or components to continue operations. Failover and recovery times vary, depending upon the application and the configuration of the reliability, availability, and serviceability (RAS) features. Only through actual testing of failover operations and for each application/failure mode can recovery times be established. Published recovery times should reflect the estimated time of failover to occur and when services are resumed.

Study the descriptions and diagrams presented in this module, examining the Sun Cluster 3.0 hardware configuration for the cluster pair. Identify each component selected and the options implemented and consider the data paths and other RAS features configured to meet the design goals for the solution; no SPOF for the cluster-pair.

Additional design decisions and configuration options are detailed during subsequent lab modules and exercises, when important availability features are implemented.

No SPOF

Multiple faults occurring within the same cluster environment can result in unplanned downtime. A SPOF can exist within the software applications architecture or a SPOF for a single node might be an embedded controller or memory module.

This basic Sun Cluster 3.0 hardware configuration is based on the Sun Enterprise™ 220R server and should be configured as an entry-level platform for providing no SPOF for the cluster pair.

Installation and Planning Considerations: General

New installations that are well-planned and well-executed are critical to ensuring reliability and ultimately, availability. Reducing the number of unplanned outages involves using proven methods when configuring HA platforms and minimizing SPOFs. Adhering to these guidelines should result in a well-executed solid installation.

The following steps can contribute to successful configurations and assist in sustaining daily operations towards maximizing platform availability and performance:

  • Document any potential SPOFs that could occur, including associated workarounds, troubleshooting procedures, and best practices.

  • Ensure that Sun Cluster 3.0 hardware administrators are highly trained and able to successfully test and conduct cluster failover operations for each HA application, and associated systems and subsystems, including fault isolation/troubleshooting and recovery procedures using all available utilities and interfaces.

  • Document site-specific and application-specific configurations and procedures as part of implementing best practices and simplifying datacenter operations and maintenance.

  • Record all standard and non-standard configurations, implementing change management procedures for all systems and sub-systems (for auditing and tracking key systems and components throughout the life cycle of the data center).

  • Implement well-known established configurations and techniques that minimize platform complexity and the number of active components to simplify operations and maintenance.

  • Provide diagnostics, utilities, and interfaces that are easy to use and interpret with clearly documented error messages and procedures for resolving potential problems.

  • Refer to the Sun Cluster 3.0 Configuration Guide for current restrictions on hardware, software, and applications running in the Sun Cluster 3.0 hardware environment.

    NOTE

    Please see ''References'' on page 19 when planning cluster installations for additional information, especially the document, Sun Cluster Site Planning.

Software Licenses

Sun Cluster 3.0 software requires its own user license. For bundled HA agents, obtain licenses from your local Sun Service provider. For HA agents developed by Sun Microsystems or third-party vendors requiring licenses, contact your local Sun Microsystems representative for professional services. Additionally, Sun Cluster 3.0 software does not include a VERITAS Volume Manager (VxVM) software or CVM software license (as Sun Cluster 2.2 software does) so it must be purchased.

NOTE

In some cases, Sun StorEdge™ arrays include a VxVM software license.

Cable Configuration Schematic

Figure 1-1 illustrates the hardware implementation and schematic for each connection.

Figure 1-1 Cable Configuration Schematic

NOTE

See Figure 1-9 for the proper stacking order of equipment.

In Figure 1-1, c1 = PCI3, c2 = PCI4 and D1000's include t0, t1, t2, t8, t9, t10.

Cable Connection Tables

Tables 1-1 through 1-5 further define the configuration schematic illustrated in Figure 1-1

Table 1-1 Server to Storage Connections

From Device

From Location

To Device

To Location

Cable Label

E220R # 1

SCSI A (PCI3)

D1000 #1

SCSI A

C3/1 - C3/3A

E220R # 2

SCSI A (PCI3)

D1000 #1

SCSI B

C3/1 - C3/3B

E220R # 1

SCSI A (PCI4)

D1000 #2

SCSI A

C3/2 - C3/3A

E220R # 2

SCSI A (PCI4)

D1000 #2

SCSI B

C3/2 - C3/3B


Table 1-2 Private Network Connections

From Device

From Location

To Device

To Location

Cable Label

E220R # 1

qfe0

E220R # 2

qfe0

C3/1 - C3/2A

E220R # 1

qfe4

E220R # 2

qfe4

C3/1 - C3/2B


Table 1-3 Public Network Connections

From Device

From Location

To Device

To Location

Cable Label

E220R # 1

hme0

Hub # 00

Port #2

C3/1 - C3/5A

E220R # 1

qfe1

Hub # 01

Port #3

C3/1 - C3/6A

E220R # 2

hme0

Hub # 01

Port #2

C3/2 - C3/5A

E220R # 2

qfe1

Hub # 00

Port #3

C3/2 - C3/6A


Table 1-4 Terminal Concentrator Connections

From Device

From Location

To Device

To Location

Cable Label

E220R # 1

Serial Port A

Terminal Concentrator

Port #2

C3/1 - C3/4A

E220R # 2

Serial Port A

Terminal Concentrator

Port #3

C3/2 - C3/4A

Terminal Concentrator

Ethernet Port

Hub # 00

Port #1

C3/4 - C3/5A


Table 1-5 Administrative Workstation Connections

From Device

From Location

To Device

To Location

Cable Label

Administration Workstation

hme0

Hub # 00

Port #4

F2/1 - C3/5A

Administration Workstation

Serial Port A

Terminal Concentrator

Port #1 **

F2/1 - C3/5B


NOTE

Here's some additional details regarding the Cable Label column in Tables 1-1 through Table 1-5:

  1. The Cable Label column assumes the equipment is located in grid location C3 (see Figure 1-8). The number following the grid location identifies the level a piece of equipment is stacked with 1 being the lowest level. For additional details please see "Step 1.2" on page 16, and Figure 1-8.

  2. The letter at the end of the label tag indicates how many cables terminate at that level. For example, the letter A indicates one cable, B indicates two cables, etc.

  3. The label tag F2 is the grid location of the administrative workstation. For additional information see "Step 1.2" on page 16.

  4. ** indicates that this cable is only connected when configuring the terminal concentrator.

Equipment Summary

The hardware used for this lab example includes, two Sun Enterprise 220R servers (cluster nodes) configured with four PCI I/O cards. Figure 1-2 shows a Sun Enterprise 220R server. The lab kit may or may not have the four I/O cards installed.

Figure 1-2 Sun Enterprise 220R Server

NOTE

Two PCI qfe I/O cards (Sun Microsystems part # 1034A) should be installed in PCI slots 1 and 2. Install the Dual-Channel Differential UltraSCSI host adaptors (Sun Microsystems part # 6541A) in PCI slots 3 and 4.

Key Practices—Specifications should ensure consistent configurations, across nodes and devices, when possible. For example, the cluster boot, boot disk, and mirror should be of the same type or model, having identical device specifications. For these lab exercises, the cluster node(s) and disk subsystem configuration are configured in a consistent, symmetric manner, as represented in Figure 1-1, and Tables 1-1 through 1-5 on page 8. As a result, configuration and sustaining operations often benefits. Ensure that the configuration of adapters and devices are consistent across all nodes. Unless the design specifications include cluster nodes that are specifically non-symmetric (due to failover and workload requirements) the components selected for use in each cluster node should be of the same type, model, or brand. During installation, ensure that all components are installed in the prescribed manner. Maintain consistent configurations for systems and subsystems. For example, if PCI slot 2 on node 1 has a PCI SCSI host adapter, then PCI slot 2 on node 2 should also have a PCI SCSI host adapter.

NOTE

The practice of ensuring consistent configurations across nodes can hold true for SBUS (EXX00) systems—if slot 0, 1, or 2 on system board 2 for node 1 has an SBUS qfe card—then the system board 2 on node 2 should have a corresponding SBUS qfe card in slot 0, 1, or 2. Following this practice assists in future troubleshooting and maintenance.

  • One 8-port Terminal Concentrator (TC).

    The terminal concentrator (see Figure 1-3) enables access to the console port on each node in the SunPlex component. Sun Cluster 3.0 hardware implementations no longer specify the TC brand and model. For some Sun Cluster 2.2 hardware implementations, the Sun Microsystems terminal concentrator was required and performed the task of failure fencing in configurations where the cluster had more than two nodes and a failure has occurred. This is not the case with a Sun Cluster 3.0 hardware setup.

    NOTE

    This component is also known as a terminal server, however, this lab guide refers to it as a terminal concentrator (TC) to avoid confusion with the typical meaning of the term server.

    Figure 1-3 Typical 8 Port Terminal Concentrator (Rear view)—Ethernet Port Circled

    NOTE

    Configuration and setup of the TC requires that the connection of a serial cable (Sun Microsystems part # 530-2152) between port 1 of the TC and the serial/tty port on the Management Server (or administrative workstation). Furthermore, it is highly recommended that the TC be implemented using the exact type specified by Sun Microsystems according to the specific brand and model. This was a requirement for Sun Cluster 2.2 hardware setup.

  • Two Sun StorEdge D1000 disk storage devices.

    The Sun StorEdge D1000 server (illustrated in Figure 1-4) has been designed to eliminate known SPOFs by incorporating dual power supplies and other redundancy features. In this exercise, each Sun StorEdge D1000 server is configured as a single backplane (loopback cables are installed for a non-split SCSI bus). Each Sun StorEdge D1000 server is configured with six disk drives, as target id 0, 1, 2, 8, 9, and 10. We use a volume manager (software RAID) to mirror across the two Sun StorEdge D1000 server arrays.

    Figure 1-4 Sun StorEdge D1000 server

  • Two 10/100 BaseT Ethernet hubs

    Figure 1-5 illustrates a typical ethernet hub.

    Figure 1-5 Typical Ethernet Hub

  • One Ultra 5™ Workstation

    Figure 1-6 shows a Ultra™ 5 workstation to be used as the administrative workstation and JumpStart™ server for the nodes. A larger workstation can also be used. Figure 1-7 shows the rear view of a Ultra 5 workstation serial port.

    Figure 1-6 Ultra 5 WorkStation

    Figure 1-7 Ultra 5 Workstation Serial Port (Rear View)—Serial Port Circled

    NOTE

    The serial cable is required when initially configuring a TC for use in the cluster.

    CAUTION

    According to common practice, when performing the hardware installation do not connect the power cords or attempt to power on the equipment until all hardware and components are installed and the cabling is completed. At the appropriate time, follow the recommendations for powering on subsystems and cluster nodes. One common practice suggests powering on all external (for example, storage) devices prior to powering on each node.

Step 1.1

For each cluster node, record the hostname, hostid, IP address, and Ethernet (MAC) address. Figure 1-8 is an illustration of equipment stacking detailing slot assignments and labels.

Figure 1-8 Equipment Stacking Showing Slot Assignments and Labels

Disk Subsystem Considerations

When configuring disk subsystem components, consider the following points, noting that clustnode2's global SCSI initiator ID is set to '6' avoiding the potential conflict with clustnode1:

  • The embedded SCSI controller (c0) connects the root/swap disk (c0t0), plus the boot mirror (c0t1). PCI slots 3 and 4 are assigned to SCSI adapters c1 and c2. PCI slot 3 (SCSI adapter) connects disks c1t0 through c1t2 and c1t8 through c1t10. PCI slot 4 (SCSI adapter) connects c2t0 through c2t2 and c2t8 through c2t10.

  • Refer to Figure 1-1 for the slot and controller assignments, SCSI target IDs, and datapaths.

  • Both cluster nodes must share access to disk array #1 and #2. This means a potential conflict will arise if both clustnode1 and clustnode2 attempt to access the same SCSI bus. Prior to enabling the two cluster nodes to share the same SCSI bus the clustnode2 global scsi-initiator-id is set to 6, while its internal assignment remains at 7. Remember that SCSI ID settings must be unique for all devices attached to the same SCSI bus, and that the SCSI ID establishes priority when multiple devices arbitrate for control of the SCSI bus. A global value of 6 establishes that clustnode2 retains high priority. It is lower only to the clustnode1 adapter(s), which remains set to the default value of 7. The settings for clustnode2 are established in the nvram, to ensure that the configuration is saved across power outages.

Additional Sun StorEdge D1000 server and disk considerations (for example, performance, availability, partitioning, etc.) are addressed during subsequent lab modules.

Private Networks

Figure 1-1 illustrates how each cluster node is interconnected via their respective qfe0 and qfe4 ports. The basic Sun Cluster 3.0 hardware platform connects two or more servers by means of a private cluster interconnect.

The cluster interconnect supports cluster application data and/or locking semantics between cluster nodes. The cluster interconnect is fundamental to cluster operations. Do not use the interconnect for routing any other traffic or data. The private network establishes exclusive use of pre-assigned or hard-coded IP addresses for each cluster node. Redundant network links (for example, fault tolerant) are implemented. Failover is transparent and immediate. The type of application determines which interconnects are supported.

NOTE

Using cross-over cables can minimize the use of hubs and conform to established Key Practices in a two-node cluster.

Terminal Concentrator (TC)

Because a cluster node does not have a monitor or keyboard, a single TC is used for console access and operations. The TC is not required for normal operation of cluster nodes. If the single TC fails, access to the cluster nodes is provided through public network connections. You may configure multiple terminal concentrators to achieve higher levels of availability.

Switches and Hubs

It is recommend that two hubs are configured in a basic cluster for redundant connections to public networks and switches. On each cluster node, Figure 1-1 on page 7 depicts how NAFO software can provide failover between hme0 and qfe1, however, Sun Professional Services strongly recommends more reliable network availability solutions using Alteon Websystem or Extreme Networks switches. See ''References'' on page 19 for additional references to specific Sun BluePrints Online articles.

TC Settings

In the space provided, record the settings and TC cable connections. This data is used to create the /etc/serialports file during subsequent lab exercises.

Port on TC

Node Name

1

clustadm

2

clustnode1

3

clustnode2

4

 


Step 1.2

One example easily identifies cables using labels. Cables used to connect equipment should be clearly marked to reflect the following information (local and manual installations only):

  • Location within a datacenter or a grid reference

  • From device including the level ID

  • To device including the level ID

The datacenter floor can be divided into a grid arrangement, similar to Figure 1-9.

Figure 1-9 Datacenter Grid Diagram

Each location within the grid is known as a cabinet. In this exercise, we have stacked the equipment at grid reference C3, with the administrative workstation located at grid reference F2. Additionally, each piece of equipment within the cabinet has a level ID. Labelling each cable end with an identification tag makes it easier for future troubleshooting or maintenance. A cable tag may look like the follow example.

Example: C3/6 - F2/1. This is from grid reference/level ID to grid reference/level ID.

Step 1.3

For manual installations, label all major components as to their function or name.

Key Practices—Label devices and cables. Devices and cables that are easily identified make troubleshooting and maintenance easier. We have found this helpful even for systems that do not require high availability. Where high availability is a pre-requisite, cable and device identification and labelling should be high priority. Verify that the disks (boot devices, mirrors, disk quorums, clones, hot-spares, etc.), the tape drives, and the cable labels are correct. This allows service operations to correlate error messages to specific global Device ID (DID) numbers, metadevice names, or the specific controller number and sd/ssd instances and help in the interpretation of these errors to determine specific failed disks and the related components. Label tape drives with their rmt instance numbers.

Step 1.4

Connect the cables to the components as illustrated in Figure 1-1 and as detailed in Tables 1-1 to 1-5 on page 8, paying special attention to the following information.

Key Practices—Connect cables used for the same function to the same cards and ports on both machines. This assists in troubleshooting; especially with installations where it is difficult to visually trace cabling between devices.

Example: If the private interconnect (heartbeat) #l is connected to port 0 of the Ethernet card in bay 1 on node #1—it should be connected to port 0 of the Ethernet card in bay 1 on node #2.

Key Practices—To minimize SPOFs, distribute the I/O and datapaths (cabling and connections) across as many different I/O Cards or system boards as possible and practical. Implementing this practice ensures that no single board failure will bring down the redundant hardware component enabling rapid recovery efforts.

Example: If the private interconnect (heartbeat) line #l is connected to the Ethernet card in slot 1 then the private network line #2 should be connected to the Ethernet card in slot 2. Therefore, if either Ethernet card fails, the other private network line is unaffected and the system continues to function.

NOTE

The private interconnect cables must be "Null" Ethernet crossover cables (Sun Microsystems part # X3837A) because they connect directly machine-to-machine. Connect qfe4 on node 1 to qfe4 on node 2, connect qfe0 on node 1 to qfe0 on node 2 as shown in Figure 1-10.

Figure 1-10 Sun Enterprise 220R Server Private Interconnect Cabling

Step 1.5

For manual installations, connect all device power cords and power on all equipment except the TC which is configured separately.

Summary of Key Practices

The following is a review list of the key practices that we have detailed so far in this lab guide.

Establish Cluster Requirements (design goals) and implement configurations to meet these specifications.

Plan the installation carefully, considering all components of the SunPlex solution and applications, throughout the production life-cycle.

Ensure consistent configurations across all nodes and devices, when possible, benefiting sustaining operations and maintenance.

Identify and label all devices and cables, clearly.

Minimize SPOFs with straight-forward designs, distributing cabling and connections across multiple system boards, and IO cards.

Configure fewer active components, in a less complex and consistent manner (for example, cross-over cables instead of active hubs).

Do not power on to any equipment until all cables and connections have been verified.

Disconnect TC port 1 when TC has been configured and verified. All access to Port 1 should be carefully managed to avoid potential security risks.


NOTE

The final validation of this module occurs during Module 2, after you have configured the TC software and have installed the Sun Cluster client software on the administrative workstation. At this time, it must be assumed that no equipment faults exist for any of the SunPlex components.

End of Module One

At the end of Module One, the hardware has been set up. In this module, you have identified each component required to successfully complete the hardware installation, including:

  1. Management Server or Admin Workstation

  2. Cluster nodes

  3. Sun StorEdge D1000 server Disk Arrays (shared storage)

  4. Private and Public Networks

  5. Terminal Concentrator (TC)

  6. Ethernet hub(s)

  7. Cabling and connections

  • + Share This
  • 🔖 Save To Your Account