Enterprises operate a rich variety of IT applications, with each having a different requirement for server hardware capabilities. Vertical industry sectors such as finance, government, energy, and telecommunications, among others, have many mission-critical applications. To enable these applications, systems must be stable and reliable, highly elastic for easy expansion, and responsive to dynamic resource allocations in a timely manner.
Huawei’s flagship KunLun servers are ideal for mission-critical applications such as large enterprise core databases, Enterprise Resource Planning (ERP), and Customer Relationship Management (CRM) systems. KunLun servers provide up to 32 Intel® Xeon E7 processors, 32 TB of Double-Data-Rate Fourth Generation (DDR4) memory, and powerful I/O expansion. The KunLun platform is built to work as a large-scale enterprise service consolidation platform for two main purposes: 1) dynamic resource allocation to improve hardware resource utilization, simplified management, and reduced operating costs and 2) isolation of discrete services to ensure reliability.
KunLun servers provide a variety of technologies and solutions, such as Physical Partitioning (K-Par) and Logical Partitioning (L-Par), as well as virtualization software and Oracle multitenant solutions to meet different service consolidation requirements. In specific scenarios, multiple technologies and solutions will be used together to further improve reliability and performance, and reduce costs. This article describes the K-Par, L-Par, and multitenant solutions in detail.
KunLun K-Par for Optimal Consolidation of Mission-Critical Applications
The K-Par feature of KunLun servers allows a server to be divided into electrically isolated partitions to provide higher levels of reliability than that can be achieved using virtualized software partitions.
With K-Par, users can allocate server resources based on real service requirements to achieve optimal performance and service reliability. K-Par also allows for high flexibility to further protect customer investments.
Figure 1: Computing chassis K-Par front view (left) and rear view (right)
KunLun K-Par has the following features:
• Easy switching: Users can change the system-working mode without changing hardware or upgrading software by mouse selection on the Central Management Console (CMC) web interface.
• Unified management: Users can manage the server configurations through the CMC interface, no matter whether in single system mode or with multiple physical partitions.
• Equal allocation of service resources: KunLun servers support physical partitions with hardware resources (CPUs, memory, and storage resources) evenly distributed.
• High reliability: The physical partitions are electrically isolated. Therefore, faults occurring in one partition will not affect any other partition. Physical partitions offer fivefold higher reliability when compared with virtual partitions.
Table 1 lists the specifications of equal physical partitions (which means each physical partition has the same resources such as CPUs). Huawei has planned the uneven physical partitioning feature for KunLun with two main enhancements. One enhancement is to add additional 12S and 20S specifications, while the other is to allow different partitions in the same cabinet to have different numbers of CPUs. Take KunLun 9016 as an example. An uneven physical partition will support one 8S physical partition and two 4S physical partitions.
Table 1: KunLun physical partition specifications
- K-Par Application Scenarios
With its high reliability and performance, the KunLun K-Par feature is ideal for mission-critical application consolidation, and especially for database consolidation.
Two KunLun servers are recommended for a typical Oracle scenario, where each KunLun machine is divided into two or more physical partitions and the physical partitions of the two servers form a Real Application Cluster (RAC).
The K-Par feature can be combined with virtualization technology to improve resource utilization and slash costs. For example, enterprise ERP databases demand high performance and the applications servers can run properly with few computing resources. In this case, a KunLun 9016 with two physical partitions can be used. The Oracle database will be deployed in one physical partition and the application servers deployed on Virtual Machines (VMs) in another partition. KunLun servers support virtualization packages such as VMware vSphere, Huawei FusionSphere, Hyper-V, and KVM. The physical partition using the virtualization software supports a maximum of eight processors.
KunLun L-Par for Efficient Service Consolidation
KunLun L-Par is designed to allow the hardware resources of each partition to be isolated, including CPUs, memory, and I/O devices. Compared with K-Par, L-Par offers greater flexibility and finer granularity. Currently, KunLun L-Par supports processors of 1 to 80 cores and memory of 1 GB to 1 TB (with a step of 64 MB). In the near future, the next KunLun L-Par version will support up to 96 cores and 4 TB memory.
L-Par uses Hypervisor firmware to isolate hardware resources. The Hypervisor firmware is a Huawei-developed virtualization platform that runs between the physical hardware and the OSs in logical partitions. The Hypervisor platform shields the physical hardware differences from the upper-layer applications and services, and provides the required hardware resources for each logical partition based on the logical partition configuration.
The logical partitions are isolated to prevent software errors and most hardware errors of one partition from affecting any other partition. Each logical partition can be powered off and maintained independently.
The Hypervisor firmware is integrated with the KunLun hardware and management software to achieve better performance and system reliability. It can also be managed with KunLun hardware in a unified manner.
- KunLun L-Par Application Scenarios
The L-Par feature can be used to consolidate mission-critical or non-mission-critical applications on UNIX servers and x86 servers; however, the logical partition specifications cannot exceed the specifications of a 4S physical partition on the KunLun server.
The L-Par feature can be deployed independently or in combination with K-Par on a KunLun server. For example, a KunLun 9008 or 9016 server can use L-Par to deploy services; a KunLun 9016 can be divided into two 8S physical partitions. Logical partitions can be deployed in the physical partition hosting SCE1-BPUA. The other physical partition can then be used to deploy services independently.
Figure 2: L-Par deployed on a KunLun server
When K-Par and L-Par are deployed concurrently, physical partitions are used to consolidate services from mid-range and high-end mission-critical servers, and logical partitions are used to consolidate applications from low-end or x86 servers. In mixed deployments, only the physical partition hosting SCE1-BPUA on the KunLun 9008 or 9016 server will support logical partitions. In the near future, the next KunLun version will support L-Par on more than two physical partitions and also allow in-service dynamic expansion of L-Par to meet more flexible service consolidation needs.
Figure 3: K-Par and L-Par co-deployed on a KunLun server
Logical partitions can be used to consolidate mission-critical and non-mission-critical applications. Mission-critical applications are usually connected to external storage arrays and deployed with database services, such as ERP. Non-mission-critical applications (for example front-end applications), generally use local storage disks and are connected to the external network through Network Interface Cards (NICs).
KunLun servers support Huawei storage devices and third-party storage, such as IBM, EMC, and HDS devices. Huawei storage devices can be used as system and data disks. The system disk is used to install the guest OSs of logical partitions. The third-party storage devices are used only as data disks.
When L-Par is used to consolidate mission-critical applications, I/O devices must be configured in redundancy mode. Two Fibre Channel (FC) cards are configured and multi-pathing software is used to ensure that a FC-card or optical-link fault will not affect access to the remote storage array. Dual-port NICs and port bonding are used to implement network redundancy. A High Availability (HA) system can be used, and the HA software, such as Veritas Cluster Server (VCS), can be deployed in logical partitions of each system to ensure reliability. The Red Hat Cluster Suite (RHCS) HA software is not currently supported.
• High performance: 1) In computing virtualization, CPU core binding is used to prevent CPU contention, improve timeliness of task processing, and meet performance requirements. 2) In I/O virtualization, L-Par adopts efficient front-end and back-end communication technology to reduce overhead caused by frequent switching of CPU modes and memory duplication. In addition, L-Par adopts hardware-assisted virtualization technology to provide PCI device pass-through features, reduce interrupts and memory duplication, and improve I/O performance in logical partitions. 3) The KunLun logical partition has a performance loss (tpmC) of about 10 percent and a CPU loss (Spec CPU) of about 5 percent when compared with the physical partition of the same specifications.
• High availability: 1) L-Par supports HA to reduce planned system downtime and improve logical partition availability. 2) The hardware resources of logical partitions are isolated so that hardware faults of a logical partition do not affect other logical partitions. 3) Logical partitions support I/O redundancy to improve I/O availability.
• High security: 1) L-Par supports CPU, memory, and I/O isolation to improve system stability. 2) Security hardening has been performed for Hypervisor to implement permission control, service tailoring, network port scan, access control, virus scan and prevention, and system risk scan and alerts.
Like physical partitions, logical partitions can be managed through the CMC Web interface. Logical partitions can also be managed through a Command-Line Interface (CLI), which provides user and network management, system upgrades, and log collection.
The logical partition management includes the following:
• Creating, deleting, reconfiguring, and querying logical partitions. A step-by-step wizard is provided to help users create or reconfigure logical partitions.
• Powering on or off, gracefully shutting down, restarting, and forcibly restarting logical partitions, and inserting or ejecting DVD-ROM discs.
• Logging in to a logical partition using VNC.
Oracle Multitenant for Database Consolidation
Today, cloud computing technologies are widely used by enterprise users. Database-as-a-Service (DBaaS) is an important part of the cloud computing infrastructure. DBaaS is designed to support a large number of users and services; and has high CPU, memory, performance, and reliability requirements.
KunLun servers are a good choice for a DBaaS platform in part because of its differentiated-memory hot-swap feature, which greatly reduces the time required for system maintenance to better meet Service Level Agreement (SLA) guarantees. KunLun servers, when combined with the multitenant feature of Oracle Database 12c, are powerful tools that can be used to consolidate multiple database servers into the cloud.
Oracle Multitenant is a new option for Oracle Database 12c Enterprise Edition that helps users reduce IT costs by simplifying consolidation, provisioning, and upgrades. It is supported by a new architecture that allows a container database to hold many pluggable databases.
Oracle Database 12c has introduced Container Database (CDB) and Pluggable Database (PDB) features. In the multitenant environment, one CDB can hold multiple PDBs. For database versions earlier than Oracle Database 12c, the instance and database are in a one-to-one or many-to-one RAC relationship. An instance can be associated with only one database, and the database can be loaded with multiple instances. The instance and database, however, cannot be in a one-to-many relationship. Oracle Database 12c enables one-to-many mapping between instance and database.
• High consolidation density: Multiple pluggable databases share memory and background processes in a single container database, which allow more pluggable databases to run on a specific platform.
• Rapid provisioning and cloning: Pluggable databases can be unplugged from one container database and plugged into another. Alternatively, you can clone a PDB within the same container database or from one container database to another.
• Unified management of multiple databases: By consolidating existing databases as pluggable databases, administrators can manage many databases as one. For example, tasks like backup and disaster recovery are performed at the CDB level.
• Improved availability and flexibility: Plugging and unplugging PDBs shortens the time that services are interrupted when migrating hardware.
• Higher security: The Oracle Database Vault, Transparent Data Encryption, Unified Auditing, and Database Firewall features can be configured at the PDB level.
Figure 4 shows the overall architecture of the KunLun Oracle Database 12c database solution. KunLun servers are ideal for all database scenarios, especially for mission-critical applications that demand high performance and reliability. In addition to high data-processing efficiency, KunLun servers provide 24/7 service continuity and stability.
Figure 4: KunLun Oracle Database 12c solution architecture
KunLun servers are ideal for Online Transaction Processing (OLTP) core database and intelligent analysis services, such as Online Analytical Processing (OLAP), data warehouse, and data mining. By leveraging the multitenant and in-memory features of Oracle Database 12c, KunLun servers can increase resource utilization, improve performance, and maximize Return on Investment (ROI).
Table 2 lists the original database information as deployed in three Oracle 11g RACs, including HR, finance, sales, production, delivery, service, and procurement information. The data is stored in conventional RAID arrays, and the storage space for different applications is physically isolated. Although this traditional deployment solution meets application requirements for servers and storage, it has drawbacks, such as low resource utilization and high management overhead.
Table 2: Databases before consolidation
The process of consolidating multiple Oracle 11g databases to the Oracle Database 12c cloud is as follows:
• Upgrade the original Oracle Database 11g software and database to Oracle Database 12c.
• Convert the original Oracle 11g databases to Non-Container Databases (Non-CDBs) of Oracle Database 12c.
• Deploy an Oracle Database 12c cluster, create CDBs, map the original storage resources to Oracle Database 12c cluster nodes, convert the Non-CDBs to PDBs using database cloning, and plug the PDBs to Oracle Database 12c CDBs.
After the consolidation, multiple PDBs share servers and storage resources, which maximizes hardware utilization. The PDBs can be unplugged and plugged across different CDBs, which facilitates changes to the SLA. Before migrating to a cloud architecture, the most important things are to define tenants and allocate computing and storage resources so that tenants will meet specified SLA-levels for applications. At the same time, tenant data must be isolated for the purposes of information security. The PDB management accounts of Oracle Database 12c are used to implement rights-based access to PDBs and self-service tenants. The Resource Manager dynamically allocates computing resources for different tenants.
With the KunLun server and Oracle multitenant feature, multiple Oracle 11g databases are consolidated as a single Oracle 12c database cloud. This consolidation improves hardware resource utilization and reduces hardware investment and software licensing fees. Most important, the consolidation allows tenant self-service, which greatly slashes management overhead.
Table 3: Comparison of resources before and after consolidation