Best Practices For Hp Bladesystem Deployments Using Hp

Transcript

Best Practices for HP BladeSystem Deployments using HP Serviceguard Solutions for HP-UX 11i May 2010 Technical white paper Table of contents Executive Summary ......................................................................................................................... 2 BladeSystem Overview .................................................................................................................... 2 Hardware Components ................................................................................................................ 2 Eliminating Single Points of Failure in HP BladeSystem Configurations.................................................... 11 Redundant Hardware Configurations within Blade Enclosures ........................................................... 12 Using Serviceguard in HP BladeSystem Configurations........................................................................ 17 Serviceguard Clustering within a Single Blade Enclosure.................................................................. 19 Clustering across Multiple Blade Enclosures or non-Blade Servers ...................................................... 20 Additional Considerations for using Serviceguard with the HP BladeSystem ............................................ 22 Conclusion .................................................................................................................................. 25 For More Information .................................................................................................................... 26 Call to Action............................................................................................................................... 26 Executive Summary HP continues to be tremendously successful in deploying server hardware consolidated into HP BladeSystem environments. The improved control of power consumption and workload management with HP Insight Dynamics – VSE software controlling the entire environment bring distinct advantages. HP Virtual Connect facilitates rapid deployment and infrastructure flexibility, reducing wiring and the effort to connect servers to network and SAN fabrics. This brings valuable benefits to customers in small, medium and large enterprises. HP Serviceguard Solutions play an important role in these environments to ensure mission-critical application availability for HP Integrity servers. Configuring highly available applications in the HP BladeSystem has some special considerations that differ from standard server rack-mount or HP Superdome deployments with HP Serviceguard. Knowledge of HP BladeSystem component placement, configuring HP Virtual Connect and an understanding of where cluster elements such as server nodes, quorum devices and storage should be located and configured within a cluster is critical to maximizing the high availability benefits of Serviceguard. The purpose of this white paper is to highlight the considerations and best practices for implementing HP BladeSystem solutions that are made highly available through the use of HP Serviceguard for HP-UX on HP Integrity BL860c and BL870c blade servers. Note the concepts of designing highly available blade configurations presented in this white paper will also apply to future generation HP Integrity blade server products when released. BladeSystem Overview The HP BladeSystem is the general name for HP's Industry Standard blade server line. It consists of a number of hardware components, software and services that are all designed to work together to provide a rack-mounted, integrated infrastructure for compute, network, storage and power elements. This section will briefly describe some of the HP BladeSystem hardware components that are the foundation for its integrated architecture, which will be used as a basis for understanding how these components can be configured to maximize server and application availability by eliminating Single Points of Failure (SPOFs) within a BladeSystem solution deployment. As you will learn in this section, many of the HP BladeSystem components have already been designed with redundancy in-mind to maximize availability and minimize downtime. Hardware Components The following are some of the major components of the HP BladeSystem. Enclosures The HP BladeSystem c-Class Enclosures are central component for joining computing resources into a consolidated, “wire-once” infrastructure. There are two c-Class enclosures available to best meet a customer’s business requirements, as shown in figure 1: • c3000 for remote sites & small to medium businesses (rack or tower configurations) • c7000 for enterprise data center applications 2 Figure 1: c-Class HP BladeSystem Enclosure Family HP BladeSystem c7000 enclosure HP BladeSystem c3000 Tower HP BladeSystem c3000 enclosure Both enclosures share common: • Half-height /full-height server blades • Interconnect modules • Mezzanine Host Bus Adapter (HBA) cards • Storage blades • Power Supplies (hot swappable and redundant) • Fans (hot-swappable and redundant) A comparison between the c3000 and c7000 enclosures is shown in Table 1. Table 1: HP BladeSystem c-Class Enclosure Comparison c3000 enclosure and tower enclosure c7000 enclosure 6U Height (rack) or tower 10U height Horizontal blade orientation Vertical blade orientation for tower Vertical blade orientation 8 HH (half-height) Blades, 4 FH (full-height) Blades,6HH/1FH 16 HH (half-height) Blades, 8 FH (full-height) Blades 4 Interconnect bays 8 Interconnect bays 6 Power Supplies @ up to 1200W each 6 Power Supplies @ up to 2250W each 6 Active Cool Fans 10 Active Cool Fans 3 Device and Interconnect Bays The interconnect bays for each enclosure can support a variety of Pass-Thru modules and switch technologies, including Ethernet, Fibre Channel, and InfiniBand. The enclosures support redundant I/O fabrics and can yield up to a 94% reduction in cables compared to traditional rack-mounted server configurations. One of the major differences between the c3000 and c7000 is in the number of available interconnect bays; the c3000 has 4 while the c7000 has 8. The four additional interconnect bays in the c7000 offer additional I/O flexibility and the ability to use redundant interconnects to eliminate single points of failure (SPOFs), which is extremely important to help protect mission-critical applications in the data center. Using redundant interconnect modules in high availability HP BladeSystem configurations will be described in later sections of this white paper. Figure 2 shows a side-view of a c7000 enclosure and the major component connections. Figure 2: HP BladeSystem c-Class Enclosure Side View Fans Half-height server blade Switch modules Half-height server blade Fans Power supply modules Signal midplane Power backplane Fans AC input module The c7000 enclosure, as with the c3000, enables easy connection of embedded server device ports from the device bays to the interconnect bays. The enclosure signal midplane transfers I/O signals (PCIe, Gigabit Ethernet, Fiber Channel) between the server blades (half-height or full-height) and the appropriate interconnects, and has redundant signal paths between servers and interconnect modules. Since the connections between the device bays (in the front of the enclosure where the blade servers reside) and the interconnect bays (in the back of the enclosure containing the interconnect modules) are hard-wired through the signal midplane, the Mezzanine cards – host bus adapters (HBAs) used to connect the blade servers with an interconnect module - must be matched to the appropriate type of interconnect module. For example, a Fiber Channel Mezzanine card must be placed in the Mezzanine connector that connects to an interconnect bay holding a Fiber Channel switch. For port mapping purposes, it does not matter in which bay you install a server blade; Mezzanine connectors in the blade expansion slots always connect to the same interconnect bays. 4 To simplify the installation of the various Mezzanine cards and interconnect modules, the Onboard Administrator, which manages the components within the enclosure, uses an “electronic keying” process to detect any mismatch between the Mezzanine cards and the interconnect modules. The power backplane provides 12V DC power to server blades, fans, and interconnects. Both the signal midplane and separate power backplane in the c7000 enclosure have no active components, thus improving reliability. The AC input module providing power to the redundant power supply modules can be configured to use a variety of different power delivery modes, depending on customer availability requirements and cost constraints. The module can be configured to use either single-phase or three-phase AC for NonRedundant Power, Power Supply Redundant and AC Redundant power delivery modes. A detailed description of these power modes and enclosure power configuration options is described in the technology brief titled “Technologies in the HP BladeSystem c7000 Enclosure” available at http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00816246/c00816246.pdf. It is recommended to use both Power Supply Redundant and AC Redundant power delivery modes to achieve the highest levels of availability, if possible. Figure 3 shows the interconnections between the device I/O from the server blades in the front of the c3000 enclosure to the interconnect switch module (SWM) ports located at the rear of the enclosure for data transfer. The color-coded symbols on the diagram are identical to the symbols used on the physical enclosure and Onboard Administrator port mapping displays to identify the interconnect bays (see figure 4). Each front device bay is connected through the signal backplane to each of the rear interconnect bays. Interconnect bays 1 (for the c3000 / c7000) and 2 (for the c7000) are dedicated to signals from the embedded NICs located on the server blade system board. The remaining interconnect bays are available to accept signals from Mezzanine HBA cards mounted directly on the server blades. The server blade Mezzanine card positions connect directly through the signal mid-plane to the interconnect bays. The interconnect bays are designed to accept single-wide or double-wide switch modules (SWMs) for interconnect bandwidth and form factor scalability. HP Integrity BL860c and BL870c blade servers are full-height and provide connections for two 2-port embedded NICs, also known as “LAN on Motherboard”, or LOM, and up to three 4-port Mezzanine HBAs (labeled Mezz-1 – Mezz-3) as shown in figures 3 and 4. PCIe (PCI Express) connectivity from the blade system board to the LOM and Mezzanine HBAs uses paired groups of full-duplex communication “lanes”. A single-wide lane provides a 1x 500MB/s transfer rate, and a double-wide or two lanes provide a 2x transfer rate of1Gb/s. Mezzanine cards are categorized into “types” that describe their data transfer capabilities. Type I Mezzanine cards provide 1x transfer rate, while Type II Mezzanine cards provide 1Gb/s through a single lane. Embedded LOM and host bus adapters installed in Mezzanine card slot 1 support single-wide lane interconnects, while Mezzanine slots 2 and 3 support either single-wide or double-wide lane interconnects. Figures 3 and 5 show the PCIe lane connections available to the LOM and Mezzanine cards on the blade server. The Integrity BL860c is a single-wide server blade. The designation “N” in the diagram is used to map single-wide server blade connections to the switch module bay ports. The BL870c is a doublewide server blade and follows a slightly different port mapping scheme in that: • If a BL870c server blade is in device bays 1 and 2, the value of "N" is 2 • If a BL870c server blade is in device bays 3 and 4, the value of "N" is 4 Several points to note regarding the c3000 diagram are: • All four LOM ports on each server blade use the same interconnect switch module bay SWM-1 • All four ports of Mezzanine card 1share the same interconnect switch module bay SWM-2 • Ports on Mezzanine cards 2 and 3 are divided between interconnect switch module bays SWM-3 and SWM-4 5 Due to the limited number of available interconnect module slots in the c3000, it is not possible to configure the enclosure for complete redundancy to eliminate a Mezzanine card and interconnect module as a single point of failure between the server blade and connectivity to the outside system infrastructure. This is an important point to consider when deploying mission-critical environments an whether this configuration will meet defined availability requirements. Figure 3: HP BladeSystem c-Class Enclosure Side View Full-Height Server Blade N (N = 1…4) 2x N+4 N+12 2x GbX1 PCIe x4 NIC 2 1 4 PCIe x4 Mezz-1 3 N 2 N N+8 N+8 N+4 N+12 1 SWM-1 SWM-2 4 PCIe x8 Mezz-2 3 2 1 PCIe x4 NIC N+4 2 1 N SWM-3 4 PCIe x8 Mezz-3 N+8 N+12 N+4 N N+8 N+12 SWM-4 3 2 1 GbX2 Blade Slot # N = 2,4 for Integrity BL870c For visual reference, Figure 4 shows the c3000 enclosure rack and tower Interconnect bay numbering scheme. 6 Figure 4: HP BladeSystem c-3000 Enclosure Rack and Tower Interconnect Bay Numbering Server blade signal Interconnect bay Interconnect bay label NIC 1, 2, 3, 4 (embedded) 1 – Orange hexagon Mezzanine 1 2 – Yellow square Mezzanine 2 3 and 4 – Green circle Mezzanine 3 3 and 4 – Blue diamond Figure 5 shows the interconnections between the server blades and interconnect switch module (SWM) ports for the c7000 enclosure, with a similar physical interconnect bay color-coding scheme (see figure 6). The mapping of the Bl860c and BL870c blade connections to the switch module bay ports is similar to the c3000 enclosure; however since the enclosure has 8 available device bays, • if a BL870c server blade is in device bays 1 and 2, the value of "N" is 2 • if a BL870c server blade is in device bays 3 and 4, the value of "N" is 4 • If a BL870c server blade is in device bays 5 and 6, the value of "N" is 6 • If a BL870c server blade is in device bays 7 and 8, the value of "N" is 8 7 Figure 5: HP BladeSystem c7000 Enclosure Interconnect Diagram 2x Full-Height Server Blade N (N = 1…8) N 2x NIC N SWM-1 GbX1 PCIe x4 N+8 N+8 SWM-2 2 1 N 4 PCIe x4 Mezz-1 N 3 N+8 N+8 2 SWM-3 1 SWM-4 4 PCIe x8 Mezz-2 3 N 2 N+8 N N+8 1 SWM-5 PCIe x4 NIC 2 1 SWM-6 N+8 N+8 N N 4 PCIe x8 Mezz-3 3 SWM-7 SWM-8 2 1 GbX2 Blade Slot # N = 2,4,6,8 for Integrity BL870c Several points to note regarding the c7000 diagram are: • The two LOM modules, each with a dedicated PCIe bus and two ports on each blade server, are divided between interconnect switch module bays SWM-1and SWM-2 (although NIC ports 1 on each LOM controller share SWM1 and NIC ports 2 share interconnect switch module SW2) • Ports on Mezzanine cards 1, 2 and 3 are divided between interconnect switch module bays SWM3 thru SWM-8 With the additional interconnect module slots in the c7000, it is now possible to configure the enclosure to eliminate both Mezzanine cards and interconnect modules as single points of failure between the server blade and connectivity to the outside system infrastructure. Therefore, deploying c7000 enclosures is a best practice recommendation for mission-critical environments. For visual reference, Figure 6 shows the c7000 enclosure Interconnect bay numbering layout. 8 Figure 6: HP BladeSystem c7000 Enclosure Interconnect Bay Numbering Server blade signal Interconnect number Interconnect bay label NIC 1 and NIC 3 ( embedded ) 1 – Orange hexagon NIC 2 and NIC 4 ( embedded ) 2 – Orange hexagon Mezzanine 1 3, 4 – Yellow square Mezzanine 2 5, 6 and then 7, 8 – Green circle/Blue diamond Mezzanine 3 7, 8 and then 5, 6 – Blue diamond/Green circle HP BladeSystem Onboard Administrator BladeSystem Onboard Administrator (OA) (figure 7) is located below the interconnect bays and provides component management in c-Class enclosures by: • Detecting component insertion and removal • Identifying components and required connectivity • Managing power and cooling • Controlling components Administrators access the BladeSystem OA in several ways: • Remotely through the web browser graphical user interface (GUI) • Scriptable command line interface (CLI) • On-site through the built-in Insight Display diagnostic LCD panel on the front of the enclosure • OA with KVM (Keyboard, Video, Mouse) module allows direct connection to the enclosure with a keyboard, video monitor, mouse or KVM switch through a VGA port When a component is inserted into a bay, the BladeSystem Onboard Administrator immediately recognizes and identifies the component through presence signals on each bay. If a component is removed from a bay, the BladeSystem Onboard Administrator deletes the information about that component from its current configuration. Each Onboard Administrator module has one Ethernet and one serial port that can be used to link enclosures in a rack. Enclosure links are designed to support only c-Class enclosures in the same rack, and both c3000 and c7000 enclosures can be linked and managed together. It is a best practice 9 recommendation to verify that both Onboard Administrator modules have the same and latest firmware revisions installed. Figure 7: HP BladeSystem Onboard Administrator HP Integrity Blade Servers HP Integrity blade servers - the BL870c and BL860c - enable customers to run and consolidate business and mission-critical applications in the flexible BladeSystem c-Class infrastructure, providing superior virtualization, high availability, scalability, simplified management, and energy efficiency. The full-height BL860c is a two-socket blade server that supports dual-core Intel Itanium 9100 processors, up to 48GB memory, four Gigabit Ethernet ports, support for 3 standard c-Class I/O mezzanine cards, and up to two internal SFF (Small Form Factor) SAS (Serial Attached SCSI) hot-plug disk drives. The BL860c is a low-cost platform suited for testing, development and production application consolidation, especially in IT organizations using HP-UX operating environments. The HP Integrity BL870c Server Blade is a four-socket, full-height double-width server blade that can support Intel Itanium 9100 series dual-core processors with up to 96GB memory, four Gigabit Ethernet ports, three standard c-Class I/O mezzanine cards, and up to four internal SFF SAS hot-plug disk drives. The BL870c is an ideal platform for use as the database tier of multi-tiered enterprises applications such as SAP, and Oracle Enterprise Applications, in addition to distributed computing applications for industries such as retail distribution, communications and financial services. Both the BL860c and BL870c blade servers support HP-UX 11i v2 and 11i v3, Serviceguard A.11.17 and later versions and can coexist with ProLiant server blades and StorageWorks storage blades within the c-Class enclosure. Table 2 shows a feature comparison of the BL860c and BL870c. 10 Table 2: HP Integrity Blade Comparison BL860c BL870c Processor Intel® Itanium® 9100 Processors 2 Sockets 1.66 GHz/18MB FSB667 1.42GHz/12MB FSB533 1.6GHz/12MB FSB533, single-core Note: 9000 series processor (Montecito) also supported Intel® Itanium® 9100 Processors 4 Sockets 1.6 GHz/24MB FSB533 1.6 GHz/18MB FSB533 1.42 GHz/12MB FSB533 Chipset hp zx2 Memory PC2-4200 DDR-SDRAM (533 MHz) 12 Sockets 48GB max (using 4GB DIMMs) PC2-4200 DDR-SDRAM (533 MHz) 24 Sockets 192GB max (using 8GB DIMMs) HDD and Controller 2 SFF Hot-Plug SAS HDDs HW RAID 1 support 4 SFF Hot-Plug SAS HDDs HW RAID 1 support Networking 4 Integrated Gigabit NICs Management Integrity iLO 2 Advanced Pack OS Support HP-UX 11i v3 and v2; OpenVMS; Linux (Red Hat and SUSE); and Windows Enclosure 8 Server Blades in c7000 4 Server Blades in c3000 4 Server Blades in c7000 2 Server Blades in c3000 Mezzanine Support 3 mezzanine cards Choices: 4Gb FC; IB; PCI-e pass-thru; 4port NIC expansion 3 mezzanine cards Choices: 4Gb FC, IB, 4-port NIC expansion Eliminating Single Points of Failure in HP BladeSystem Configurations Designing a system architecture that eliminates all hardware-related single points of failure is the first step in achieving high availability for business and mission-critical applications. While there are many components of the HP BladeSystem that are redundant, it is still possible to configure HP BladeSystem solutions that have single points of failure (SPOFs). It is up to the system architect to carefully design HP BladeSystem solutions that will mitigate these SPOFs before using Serviceguard to achieve the highest levels of availability. This section will describe where SPOFs can occur in HP BladeSystem configurations and how to avoid them. Once these SPOFs have been addressed from a hardware- 11 perspective, the next section of this white paper will describe how to incorporate Serviceguard into the solution to maximize overall availability. Redundant Hardware Configurations within Blade Enclosures For data center applications, the HP BladeSystem c7000 enclosure includes the following redundant components: • Up to 6 power supplies • Up to 10 Active Cool fan kits • Up to 8 interconnect modules • Redundant signal midplane paths • Redundant enclosure management using Active / Standby Onboard Administrators (OAs) HP Integrity Server Blade redundancy features include: • Two dual-port Gigabit Ethernet embedded NICs (LOM) • 3 Mezzanine I/O slots • Up to two internal SFF (Small Form Factor) SAS (Serial Attached SCSI) hot-plug hard drives (BL860c) • Up to four internal SFF SAS hot-plug disk drives (BL870c) The following sections describe how to utilize many of these redundancy features in HP BladeSystem solution designs. Redundant Enclosure Management Redundant enclosure management is an optional feature of both the c3000 and c7000. Utilizing this feature requires installing a second HP BladeSystem Onboard Administrator (OA) module in the enclosure slot adjacent to the first OA to serve as a completely redundant controller in an activestandby mode. The redundancy logic of the OA module pair is based on a continuous heartbeat between the two modules over a dedicated serial connection in the signal midplane. If the period between heartbeats exceeds an internal timeout value set within the OA software, the standby module automatically takes control of the enclosure and becomes the active BladeSystem Onboard Administrator. When two Onboard Administrator modules are installed in the enclosure, either module can be the active module with the other becoming the standby module. Configuration data is constantly replicated from the active Onboard Administrator module to the standby Onboard Administrator module, regardless of the bay in which the active module currently resides. If two Onboard Administrator modules of the same firmware revision are installed, the module on the left of the enclosure will be the active OA. If two Onboard Administrator modules installed into the same enclosure have different firmware versions, the automatic configuration synchronization feature is disabled. When the active Onboard Administrator module fails, the standby Onboard Administrator module automatically becomes active. This occurs regardless of the position of the active Onboard Administrator module. This automatic failover occurs only when the currently active module comes completely offline and the standby module can no longer communicate with it. In all other cases, a system administrator must initiate the failover by logging into the standby module and promoting it to be the active Onboard Administrator. After the failed Onboard Administrator module is replaced (the modules are hot-pluggable), it automatically becomes the standby module and receives the 12 configuration information from the active module. It remains as the standby until either a system administrator manually promotes it to the active module or the active module fails. General HP BladeSystem I/O Redundancy Considerations HP Integrity server blades I/O connectivity provides for: • 2 LAN on Motherboard (LOM) modules (2 ports each @ 1Gb/s; 4 ports total) • Support for up to 3 Mezzanine cards With this hardware connectivity, it is possible to use a combination of LOM ports and Mezzanine cards to provide redundancy for both network and storage connections to the server blade. HP recommends configuring primary and alternate paths to use different Mezzanine cards and interconnect modules to eliminate these components as potential SPOFs, if possible. However; depending on the Mezzanine card configuration chosen based on customer availability requirements and cost, it is acceptable to have both primary and alternate paths defined through one multi-port Ethernet or Fibre Channel Mezzanine card. Networking Connectivity With the four internal Ethernet ports provided on Integrity server blades using two LOM NICs that have independent hardware port controllers, it is possible to create a redundant networking configuration that eliminates the blade networking ports as a SPOF. This is achieved by configuring the LOM ports to avoid a potential port controller failure that could disable two ports by using ports 1 and 4 as the Serviceguard primary and standby connection for one network (e.g., site data LAN) and ports 2 and 3 as the primary and standby connection for another network (e.g., Serviceguard heartbeat). Using this configuration, Serviceguard local LAN failover would protect against either a port controller or interconnect module failure. Note that it is also possible to use APA (Auto-Port Aggregation) LAN_MONITOR mode to provide an active / standby network port configuration. However; APA trunking or load balancing is not supported with Virtual Connect as Virtual Connect does not provide pass-through of LACP (Link Aggregation Control Protocol) frames to host systems. HP also recommends using an additional 4-port Ethernet Mezzanine card, if required, to provide additional network connections based on application use requirements (e.g., VM host supporting multiple VM networks). Using redundant HP Virtual Connect (VC) Ethernet modules is another method to improve network connection availability. VC Ethernet modules, when installed in a side-by-side bay pair configuration in interconnect bays 1 and 2, run as a high availability pair. Redundancy daemons running on both modules determine the active VC Manager (usually in bay 1) using internal heartbeats maintained over multiple paths (signal midplane, Ethernet link, Onboard Administrator) and can automatically switch to the other VC Ethernet module in the event of a loss of heartbeat. There are no specific network requirements for using Serviceguard with Virtual Connect other than the recommendation to eliminate a SPOF by using redundant VC Ethernet modules. Virtual Connect also facilitates Ethernet link failover by allowing Virtual Connect networks to utilize ports on multiple Virtual Connect modules in the same VC Domain. VC domains using Virtual Connect Manager can span up to four enclosures; additional enclosures can be managed using Virtual Connect Enterprise Manager. Depending on the configuration, a VC network will transparently shift its upstream communication to a port on the same module or on a different module in the event of a link failure. HP recommends using fully redundant interconnection of Virtual Connect Ethernet modules so that, if a stacking cable is lost, Ethernet packets within the VC domain will be automatically rerouted to the uplink through the redundant path. This connection also preserves network connectivity if an Ethernet interconnect module fails or is removed. Figure 8 shows an example of stacked Virtual Connect Ethernet modules. 13 Figure 8: Example of Stacked Virtual Connect Ethernet Modules Fiber Channel SAN Connectivity 2-port Fiber Channel Mezzanine cards are available for connecting server blades to a SAN infrastructure. HP recommends using two FC Mezzanine cards for redundancy to eliminate the Mezzanine cards as a SPOF. When using Virtual Connect Fiber Channel modules, HP recommends deploying the modules as side-by-side interconnect bay pairs for module redundancy with each Fiber Channel port. Note that Virtual Connect Fiber Channel modules do not have any interdependencies or mechanisms within the modules themselves to support VC-FC module failure failover as do Virtual Connect Ethernet modules. Multi-pathing for Fiber Channel links can be provided by using HP-UX 11i v3 native multi-pathing, LVM PV (Physical Volume) Links or VERITAS DMP (Dynamic Multi-Pathing). Serviceguard monitoring and failover triggered by a failed Fiber Channel link can be accomplished by using the EMS Disk Monitor for LVM by configuring a package dependency on EMS disk monitor, or by using the VxVM Volume Monitor that was available starting with Serviceguard A.11.18. Information on using EMS Monitors is available at http://www.docs.hp.com/en/B5735-90001/ch01s04.html and rules for using the HA Disk Monitor with Serviceguard is available at http://www.docs.hp.com/en/B573690074/ch02s02.html . The VxVM Volume Monitor is documented in the Managing Serviceguard Manual, which is available at http://docs.hp.com/en/B3936-90140/B3936-90140.pdf. HP Virtual Connect (VC) High Availability Server Profile Configurations Instead of having to physically wire LAN and SAN connections to specific blade servers, HP Virtual Connect (VC) server profiles provide the ability to virtually “wire” individual LAN and Fibre Channel ports to specific networks and SANs by associating multiple network ports from one or more server blades with one or more external output ports on a VC Ethernet card. From a Serviceguardperspective, if both the primary and standby LANs for a server blade shared the same output port on the same Virtual Connect module, there would only be protection against a failure of the network port on the Mezzanine card. A failure of the Virtual Connect module or the external switch connected to the output port on that module would still be a SPOF. 14 An improvement of this configuration would be to ensure the primary and standby LANs pass through different VC modules, with the output ports on the VC modules connected to separate Ethernet switches that are bridged together. This configuration would protect against the failure of a LOM port controller, the network ports on the Mezzanine card, a VC module, a switch or cabling, and would eliminate many more possible failure points. An example of such a configuration, using LOM ports 1 and 4 routed to different VC modules, is described in the next section. The following is one example of using HP Virtual Connect to create a server profile for a BL860c server blade to minimize SPOFs for its network and Fiber Channel connections. Note this is one example, and many other acceptable HA configurations are possible depending on what Mezzanine cards and interconnect modules are available for a server blade to utilize. Figure 9 shows an HP BladeSystem Onboard Administrator window with information on a BL860c server bladed installed in device bay 7. Figure 9: Onboard Administrator Window showing BL860c information for Device Bay 7 LOM and mezzanine card information In this example, the BL860c has 4 embedded LOM NIC ports (ports 1 and 2 sharing one PCIe bus and ports 3 and 4 sharing another PCIe bus), an iLO (Integrated Lights-Out) port, a 4Gb Fiber Channel Mezzanine card in Mezzanine Slot 1 and a Quad Port 1Gb NIC Mezzanine Card in Mezzanine Slot 2. Figure 10 shows the Port Mapping view for this server blade in Device Bay 7. 15 Figure 10: Onboard Administrator Window showing Device Bay 7 Port Mapping From the Port Mapping view, the embedded network ports, FC HBA ports from Mezzanine Slot 1 and Ethernet ports from Mezzanine Slot 2 are shown mapped to their respective interconnect bay ports. Note in this example configuration, ports 3 and 4 of the Quad-Port Mezzanine HBA are not mapped because there are no interconnect modules in enclosure interconnect bays 7 and 8. Although these ports are unavailable, the LOM ports on the blade server provide network redundancy, while at least two ports of the Quad-Port Mezzanine card can provide additional network connectivity. To utilize the redundant components available for the server blade in this configuration example, a Virtual Connect server profile has been created as shown in figure 11. The server profile has been configured to use LOM port 1 for the primary site LAN and LOM port 4 for the private Serviceguard heartbeat LAN, with are assigned to interconnect bays 1 and 2, respectively. Note that LOM ports 2 and 3 were not used in this case because the standby site and secondary heartbeat LAN connections have been configured to use the Ethernet ports on the Mezzanine Card Slot 2, which are assigned to interconnect bays 5 and 6. If no additional Ethernet Mezzanine cards were available, all 4 internal ports could have been used to achieve a redundant network configuration by using ports 1 and 4 as the primary/standby pair for the site network and ports 2 and 3 as primary/standby pair for the Serviceguard heartbeat network. This configuration now provides Ethernet hardware redundancy for both the site and Serviceguard heartbeat LANs. In the event of either a LOM, Ethernet Mezzanine card, or interconnect module failure, Serviceguard can recover from the failure by performing a local LAN failover. Since this configuration only has one 2-port Fiber Channel Mezzanine HBA installed in Mezzanine Slot 1, it is only possible to configure the server profile to use each port assigned to interconnect bays 3 and 4. Although a 2nd Fiber Channel Mezzanine card is not available, the ports of the one Fiber Channel Mezzanine card are using separate interconnect modules, thus eliminating the interconnect modules as a potential SPOF. However; for the highest level of availability, it is a recommended best practice to use two Fiber Channel Mezzanine cards for compete redundancy. 16 Figure 11: Server Profile for BL860c installed in Device Bay 7 Primary site LAN and private heartbeat LAN assigned to Interconnect Bays 1 & 2 Standby site LAN and private heartbeat LAN assigned to Interconnect Bays 5 & 6 Fiber Channel Ports 1 and 2 assigned to Interconnect Bays 3 & 4 In summary, this example uses the following Virtual Connect server profile configuration to minimize SPOFs with the components that were available to this BL860c server blade: • LAN on Motherboard (LOM): – Port1: Site LAN – Port 2: (unassigned) – Port 3: (unassigned) – Port 4: Serviceguard Heartbeat LAN • Mezzanine Card 1 (Fiber Channel HBA) – Port 1: SAN (primary) – Port 2: SAN (alternate) • Mezzanine Card 2 (Ethernet HBA): – Port 5: Site LAN (standby) – Port 6: Serviceguard Heartbeat LAN (secondary) • Mezzanine Card 3 (not installed – recommend using a 2nd Fibre Channel HBA for redundancy) Using Serviceguard in HP BladeSystem Configurations With the existing high availability features of the HP BladeSystem, adding Serviceguard for HP-UX on Integrity blades builds upon that foundation to provide improved availability for mission-critical applications. The complete Serviceguard Solution portfolio, including the Serviceguard Storage Management Suite, SGeRAC, SGeSAP, Enterprise Cluster Master Toolkit and all Serviceguard 17 Disaster Recovery solutions (i.e., Extended Distance Serviceguard clusters, Metrocluster, Continentalclusters) are fully supported with the HP BladeSystem BL860c (A and B versions) and BL870c Integrity server blades using the following HP-UX operating systems and Serviceguard versions: • HP-UX 11i v2 September 2006 (or later) with Serviceguard A.11.17 and SGeRAC A.11.17 (or later) • HP-UX 11i v3 September 2007 (or later) with Serviceguard A.11.18 and SGeRAC A.11.18 (or later) The following Mezzanine host bus adapters (HBAs) for the BL860c / BL870c Itanium server blades supported with Serviceguard at the time of this white paper’s publication. Please contact your local HP Sales Representative for a list of all currently supported HBAs with Serviceguard. • Ethernet: – HP BLc NC360m 2-Port Gigabit Ethernet adapter (P/N 445978-B21) – HP BLc NC364m 4-port Gigabit Ethernet adapter (P/N 447883-B21) • Fibre Channel: – HP BLc QLogic QMH2462 2-port 4Gb FC HBA (P/N 403619-B21) – HP BLc Emulex LPe1105 2-port 4Gb FC HBA (P/N 403621-B21) – HP BLc QLogic QMH2562 2-port 8Gb FC HBA (P/N 451871-B21) – HP BLc Emulex LPe1205 2-port 8Gb FC HBA (P/N 456972-B21) • SAS (Serial Attached SCSI): – HP Smart Array P700m/512 Controller (P/N 508226-B21) • InfiniBand: – HP BLc 4X DDR InfiniBand HBA (P/N 410533-B21) – HP BLc 4X DDR Dual Port InfiniBand HBA (P/N 448262-B21) The following interconnect modules for the BL860c / BL870c Itanium server blades and c-Class enclosures are currently supported with Serviceguard: • Ethernet Interconnects: – HP BLc 1Gb Ethernet Pass-thru Module (P/N 406740-B21) – HP BLc GbE2c Ethernet Blade Switch (P/N 410917-B21) – Cisco Catalyst Blade Switch 3020 (P/N 410916-B21) – Cisco Catalyst Blade Switch 3120G (P/N 451438-B21) – Cisco Catalyst Blade Switch 3120X (P/N 451439-B21) – HP BLc 1/10Gb-F Virtual Connect Ethernet Module (P/N 447047-B21) – HP BLc 1/10Gb Virtual Connect Ethernet Module (P/N 399593-B22) – HP ProCurve 6120XG Ethernet Blade Switch (P/N 516733-B21) – HP ProCurve 6120G/XG Ethernet Blade Switch (P/N 498358-B21) – HP 10GbE Ethernet Pass-thru Module (P/N 538113-B21) – HP 1:10Gb Ethernet BL-c Switch (P/N 438031-B21) • Fibre Channel Interconnects: 18 – HP BLc 4Gb Fibre Channel Pass-thru Module (P/N 403626-B21) – Brocade 4/12 SAN Switch (P/N AE370A - Note P/N AE373A to upgrade the AE370 12 port switch to 24 ports is also supported) – Brocade 4/24 SAN Switch for c-Class BladeSystem (P/N AE372A) – Cisco MDS 9124e 12-port Fabric Switch (P/N AG641A) – Cisco MDS 9124e 24-port Fabric Switch (P/N AG642A) – HP B-series 8/12c SAN Switch (P/N AJ820A) – HP B-series 8/24c SAN Switch (P/N AJ821A or AJ822A) – HP BLc 4Gb Virtual Connect Fibre Channel Module (P/N 409513-B21 or 409513-B22) – HP Virtual Connect 8Gb 24-Port Fibre Channel Module (P/N 466482-B21) – HP Virtual Connect 8Gb 20-Port Fibre Channel Module (P/N 572018-B21) • SAS Interconnects: – HP 3Gb SAS BL Switch (P/Ns AJ864A or AJ865A; note this interconnect module is supported only with the P700m SAS Controller mezzanine HBA) • InfiniBand Interconnects: – HP BLc 4X DDR InfiniBand Switch Module (P/N 410398-B21) – HP BLc 4X DDR InfiniBand Gen 2 Switch Module (P/N 489183-B21) Serviceguard Clustering within a Single Blade Enclosure With the ability of HP BladeSystem enclosures to contain multiple HP Integrity server blades, it is possible to run a Serviceguard cluster completely self-contained within an enclosure. Figure 12 shows an example of this type of configuration, also known as a “cluster in a box”. Figure 12: Example of a Serviceguard “cluster in a box” Configuration using a c7000 Enclosure HP Systems Insight Manager (SIM) Central Management Server (CMS) Redundant Network Links Quorum Service (Linux OS) Serviceguard cluster c7000 BladeSystem Enclosure Redundant Fibre Channel Links EVA Disk Array 19 In this example, a 4-node Serviceguard cluster is configured in a single c7000 enclosure, with an EVA disk array used for shared storage between the cluster nodes. An HP Systems Insight Manager Central Management Server is also shown, which provides overall management of the systems environment from outside of the Serviceguard cluster. While this example shows 4 integrity server blades used as cluster nodes, it is also possible to use HP Integrity Virtual Machines as Serviceguard nodes. HP Virtual Connect (not shown in this figure) can be configured to provide a private cluster heartbeat network within the enclosure for the cluster nodes without requiring any external wiring or switches. A quorum service, running on a Linux OS in this example, provides quorum for the 4-node cluster. Note it is supported to use a cluster lock disk or lock LUN for a 2, 3 or 4-node configuration within a blade enclosure; however it is recommended to use a quorum service for clusters having 3 or mode nodes. While this configuration is supported, it is not recommended because the blade enclosure is considered a single point of failure (SPOF) that could potentially fail and bring down the entire cluster. However; one recommended best practice shown in this diagram is the placement of the CMS on a system external to the blade enclosure so that it can remain functional for managing other systems in the environment in the event the blade enclosure is unavailable due to some firmware update operations requiring the entire enclosure to be down or a power failure of the enclosure. Advantages and Limitations This configuration has the following advantages and limitations: Advantages: • Provides a completely self-contained Serviceguard cluster within a single enclosure • Internal cluster heartbeat network can be configured using Virtual Connect to eliminate additional network cabling and switches • Provides consistent management of server profiles using Virtual Connect with all cluster nodes within the blade enclosure Limitations: • The blade enclosure is a single point of failure that can cause the entire cluster to go down • There are no nodes external to the cluster to failover workloads in the event of planned enclosure maintenance (e.g., Virtual Connect and / or Onboard Administrator firmware upgrades that require all blades in the enclosures to be shutdown) Clustering across Multiple Blade Enclosures or non-Blade Servers One architecture design for improving a “cluster in a box” configuration is to split the Serviceguard cluster nodes between multiple blade enclosures or other external Serviceguard cluster nodes to avoid having a single enclosure as a single point of failure (SPOF). Figure 13 is an example of this architecture with a Serviceguard cluster spanning multiple c7000 blade enclosures. 20 Figure 13: Cluster Example Spanning Multiple c7000 Enclosures HP Systems Insight Manager (SIM) Central Management Server (CMS) Quorum Service (HP-UX OS) LAN VC Stacking Link Serviceguard cluster SAN c7000 BladeSystem Enclosure c7000 BladeSystem Enclosure EVA Disk Array In this example, a single 8-node Serviceguard cluster spans two c7000 enclosures, with 4 nodes in each enclosure attached to shared storage provided by an EVA disk array. A Virtual Connect stacking link is used between the enclosures to provide a private cluster heartbeat network between the enclosures. A Systems Insight Manager Central Management Server is used to provide overall management of the systems environment; however this server is not part of the Serviceguard cluster. Note that it is permissible to mix server blades with other external physical or virtual (i.e., nPar, vPar, HP Integrity Virtual Machine) Serviceguard nodes in this configuration. However; no enclosure or complex can contain more than half of the cluster nodes (e.g., server blades, nPars), and requires an external quorum server to maintain cluster quorum in the event of a blade enclosure failure. A quorum service, running on a small HP-UX system in this example, is located outside of the blade enclosures to serve as a tie-breaker in case of an enclosure failure to allow the remaining 50% of the surviving cluster nodes to successfully form a new cluster. Additional information on Serviceguard cluster quorum requirements is available in the white paper titled “HP Serviceguard Cluster Configuration for HP-UX 11i or Linux Partitioned Systems” posted at http://docs.hp.com/en/6033/HPServiceguardClusterConfig_WP.pdf. Advantages and Other Considerations Having a Serviceguard cluster span multiple c7000 enclosures or other non-blade cluster nodes has many advantages over a “cluster in a box” configuration, and few limitations: 21 Advantages: • Protects against a complete blade enclosure failure • Provides the flexibility of moving workloads to another enclosure for planned maintenance (e.g., Virtual Connect and / or Onboard Administrator firmware upgrades that require all blades in the enclosures to be shutdown) • Internal cluster heartbeat network can be configured using Virtual Connect stacking links connected between the Virtual Connect interconnect modules and different enclosures to eliminate additional network cabling and switches • Provides consistent management of server profiles using Virtual Connect or Virtual Connect Enterprise Manager when all cluster nodes are in blade enclosures Other Considerations: • Additional cost for a 2nd blade enclosure or other non-blade cluster nodes • A quorum server outside of the enclosure is required for this configuration • Configurations must have exactly half of the cluster nodes in an enclosure, complex (e.g., nPar, vPar) and/or other combination of external servers Additional Considerations for using Serviceguard with the HP BladeSystem The following is a list of the major recommendations covered in this white paper and additional points to consider when configuring HP BladeSystem solutions with Serviceguard to maximize application availability: HP BladeSystem Enclosures and Internal Storage: • Ensure the firmware versions on all blade servers, Mezzanine cards, interconnect cards and Onboard Administrators are consistent and current within an enclosure participating in the Serviceguard cluster and VC domain to ensure reliable application failover • Since some Virtual Connect and Onboard Administrator firmware upgrades require all blades within the enclosure to be down, it is recommended to configure Serviceguard clusters between enclosures and failover applications before performing enclosure firmware upgrades or other planned enclosure maintenance • The SB40c Storage Blade is not supported for use as Serviceguard shared storage • The Internal disks in a server blade cannot be used for Serviceguard shared storage; however they can be used as boot/root disks with either MirrorDisk/UX or the embedded RAID controller for data redundancy Ethernet Connectivity: • The 10/100 Base-T iLO port on the c-Class BladeSystem Enclosure cannot be used in the Serviceguard cluster configuration or with Serviceguard relocatable IP addresses • LAN on Motherboard (LOM) ports are supported with Serviceguard; however for Serviceguard to reliably detect a LOM port failure on BL860c “A” version and BL870c blades, the Serviceguard cluster configuration must have the NETWORK_FAILURE_DETECTION parameter set to INONLY_OR_INOUT (the default for this parameter is INOUT) 22 – For more information on setting this parameter, see the Serviceguard Network Manager Inbound Failure Detection white paper at: http://docs.hp.com/en/5568/serviceguard.network.manager.pdf – This issue also affects HP-UX APA (Auto-Port Aggregation) link aggregates and APA failover groups (LAN_MONITOR mode) – This hardware issue and a list of available solutions is documented in the HP Support Communication Customer Advisory at: http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01814615&lang=en&cc =us&taskId=101&prodSeriesId=3676868&prodTypeId=3709945 • Recommend using network ports from different LOM port controllers for redundant active/standby Serviceguard site and heartbeat network configurations Fibre Channel Connectivity: • Each server blade must have at least 1 Mezzanine card for Fiber Channel connectivity – If a server blade has 2 or more Fiber Channel Mezzanine cards, it is recommended that the primary and alternate Fiber Channel paths use different Mezzanine cards to prevent a Mezzanine card from being a single point of failure Virtual Connect: • Uplinks between Virtual Connect modules can be used for private cluster heartbeat networks between blade enclosures (note stacking links between adjacent VC Ethernet modules is internal) – Note when using dedicated internal cluster heartbeat networks, it can be difficult to test cluster failover due to a loss of heartbeat as there is no physical LAN cable available to disconnect Flex-10: • The HP NC532m Dual Port 10GbE BL-c Adapter is supported with HP-UX and Serviceguard • Can use the 1Gb Ethernet Mezzanine or LOM with the HP BLc Virtual Connect Flex-10 10Gb Ethernet interconnect module (figure 14) with server blades running either HP-UX or Windows The Flex-10 module will only operate as a 1Gb, non-Flex-10 module for HP-UX. While it cannot take advantage of some Flex-10 features; it can be used at 1Gb and is supported with HP-UX and Serviceguard. Figure 14: HP BLc Virtual Connect Flex-10 10Gb Ethernet Interconnect Module 23 InfiniBand: • The HP BLc 4X DDR InfiniBand Mezzanine card, which requires the HP BLc 4X DDR IB Switch Module (figure 15), is supported with Serviceguard Figure 15: HP BLc 4X DDR InfiniBand Mezzanine card and HP BLc 4X DDR IB Switch Module • Considerations for InfiniBand use: – Few applications use native InfiniBand protocol; thus requiring the use of IPoverIB protocol (e.g., Oracle RAC 10g and 11g currently support only IPoverIB), which dramatically increases CPU overhead – If VERITAS CVM or CFS is used, InfiniBand must not be configured as the Serviceguard cluster heartbeat – Using InfiniBand limits the ability to have high availability configurations for the Fibre Channel and Ethernet mezzanine card as the IB interconnect module physically requires two interconnect bay slots Serviceguard Solutions Portfolio: • The Serviceguard Storage Management Suite, SGeRAC, SGeSAP, and Enterprise Cluster Master Toolkit can be used with HP BladeSystem configurations without any constraints or special considerations – Follow published manuals, release notes and white papers for suggested best practice configurations • HP BladeSystems are supported with all Serviceguard Disaster Recovery solutions (i.e., Extended Distance Serviceguard clusters, Metrocluster, Continentalclusters) Other Areas to Improve HP BladeSystem Solution Availability: • Consider adding high availability to the Central Management Server. See the white paper titled “Using HP Insight Software from a Highly Available Central Management Server with Microsoft Cluster Service” posted at http://h20195.www2.hp.com/V2/GetPDF.aspx/c01956953.pdf for more information. • Also consider configuring the quorum service as a high availability Serviceguard cluster, which is described in the HP Quorum Server documentation posted at http://docs.hp.com/en/ha.html#Quorum%20Server 24 Conclusion The HP BladeSystem has many redundant features within its design to make it highly available. Serviceguard for HP-UX on Integrity blades builds upon the HA features of the HP BladeSystem and improves availability for mission-critical applications by fully utilizing its robust feature set that can: • Detect hardware and software failures, and automatically moving critical applications to another cluster node to minimize application downtime • Integrate with HP’s partitioning solutions to provide protection for system configurations that best meet customer’s needs • Deliver application-targeted availability through the use of Serviceguard extensions for Oracle RAC and SAP, Storage Management, Toolkits and the Developers Toolbox • Provide a variety of disaster recovery options within the Serviceguard solutions portfolio There are several key points to consider when implementing Serviceguard with HP BladeSystem configurations to architect a solution that will maximize high availability: • Configure blade components (e.g., LAN on motherboard, mezzanine HBAs and interconnect modules) with redundancy in-mind within blade enclosures to avoid SPOFs (note many configurations are possible) • Consider potential cluster partitioning issues when configuring Serviceguard clusters that span multiple enclosures or use external server nodes • Consider the complete systems environment and look for other areas to improve overall availability (e.g., CMS & quorum service clustering) While the focus of this white paper was on HP Integrity server blades and Serviceguard solutions, please note that the HP BladeSystem supports both HP Integrity and ProLiant server blades. Integrated solutions for protecting HP BladeSystem continuity of services are also available for both server platforms using virtual logical servers with Virtual Connect and Virtual Connect Enterprise Manager, which are part of the Insight Dynamics advanced infrastructure lifecycle management software. Additional protection is available for each of these platforms, as listed below: HP Integrity: • Mission critical “Application-aware” availability and disaster recovery is provided using the Serviceguard family of products • Movement of server profiles is available using HP Insight Dynamics - VSE for HP Integrity servers HP ProLiant: • Availability and disaster recovery can be provided using logical server profile recovery with Insight Dynamics recovery management and HP Virtual Connect Enterprise Manager integrated with HP Insight Dynamics suite for ProLiant servers Please see the HP Insight Dynamics Protect Continuity of Services web page at http://h18004.www1.hp.com/products/solutions/insightdynamics/protect.html for details on these solutions. 25 For More Information To read more, see: • HP BladeSystem: http://www.hp.com/go/bladesystem • HP Serviceguard Solutions: http://www.hp.com/go/serviceguardsolutions • HP Insight Dynamics: http://www.hp.com/go/insightdynamics • HP Insight Dynamics – VSE for Integrity servers: http://www.hp.com/go/vse Call to Action HP welcomes your input. Please give us comments about this white paper, or suggestions for LVM or related documentation, through our technical documentation feedback website: http://docs.hp.com/en/feedback.html Share with colleagues © Copyright 2010 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice. The only warranties for HP products and services are set forth in the express warranty statements accompanying such products and services. Nothing herein should be construed as constituting an additional warranty. HP shall not be liable for technical or editorial errors or omissions contained herein. Trademark acknowledgments, if needed. HP Publication Number 5697-0470, May 2010