An Oracle White Paper October Oracle's SPARC M5-32 and SPARC M6-32 Servers: Domaining Best Practices - PDF

Please download to get full document.

View again

of 23
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Leadership & Management

Published:

Views: 4 | Pages: 23

Extension: PDF | Download: 0

Share
Related documents
Description
An Oracle White Paper October 2013 Oracle's SPARC M5-32 and SPARC M6-32 Servers: Domaining Best Practices Introduction... 1 Why Server and Application Consolidation?... 2 Requirements for Consolidation...
Transcript
An Oracle White Paper October 2013 Oracle's SPARC M5-32 and SPARC M6-32 Servers: Domaining Best Practices Introduction... 1 Why Server and Application Consolidation?... 2 Requirements for Consolidation... 3 Consolidation on Large, Vertically Scalable SMP Servers... 3 Vertically Scalable High-End SMP Servers... 4 SPARC M5-32 and SPARC M6-32 Server Consolidation Technologies 4 Dynamic Domains... 5 Oracle VM Server for SPARC... 6 Oracle Solaris... 7 Oracle Solaris Zones... 7 Oracle Solaris Resource Manager... 7 Management of the Consolidation Technologies... 8 Layered Consolidation with SPARC M5-32 and SPARC M6-32 Servers 8 A Consolidation Philosophy... 9 Dynamic System Domains (PDoms) PDom Sizing: Implications on Performance, Availability, and Flexibility 12 Oracle VM Server for SPARC LDoms Inside Dynamic Domains Guest Root Domains Zones Use Cases Conclusion Summary Appendix: Oracle SuperCluster M6-32 Configuration Rules Oracle SuperCluster M6-32 Domain Building Blocks PDom Configuration: Base or Extended LDom Configuration: One to Four LDoms Oracle SuperCluster M6-32 Conclusion... 24 Introduction The benefits of enterprise consolidation are well understood. By consolidating workloads, applications, databases, operating system instances, and servers, it is possible to reduce the number of resources under management, resulting in improved system utilization rates and lower costs. With higher utilization rates, the need to make additional hardware purchases is reduced. If consolidation also can be combined with simplification of the overall IT infrastructure, considerable savings can be made in the operational costs of running the data center. Consolidation also contributes to strategic goals, such as improving security, delivering more predictable service levels, and increasing application deployment flexibility. With the addition of the SPARC M6-32 into Oracle s server product line, price/performance scales linearly without the cost penalty for Big Iron or its enhanced features. What this means is that 16 SPARC T5-2s with 32 total CPUs are priced similarly to a SPARC M6-32 with 32 CPUs. This effectively removes the large price premium traditionally associated with this class of system, giving users an additional reason to use bigger servers: Namely, that for the SPARC platform it is no longer cheaper to procure a number of smaller servers instead of a single larger one. For successful consolidation deployments, it is necessary to select a server platform that has the scalability to support many application instances. Additionally, the server platform must have the high availability needed for mission-critical applications, the resource management and virtualization capabilities to simplify managing numerous applications, and the tools to manage the consolidated environment. Oracle s SPARC M5-32 and SPARC M6-32 servers deliver on all these requirements and are ideal solutions for server consolidation. With the SPARC M5-32 and SPARC M6-32 servers, IT managers can create pools of compute resources that can be rapidly and dynamically allocated to meet new and changing workloads. 1 Why Server and Application Consolidation? Traditionally, applications have been deployed on a single server for each application instance. In the case of complex enterprise applications, this style of deployment means that data centers require many servers for a single application, with separate servers for the web tier, application tier, and database tier. Furthermore, many enterprise applications require test and development servers in addition to the production servers. Commonly, the production servers, when initially deployed, have enough headroom to support spikes in the workload, but as the applications grow, the only way to add more capacity is to add more servers, thereby increasing complexity. As the number of servers increases, the number of operating system (OS) instances that need to be managed also grows, adding further layers of complexity and reducing IT flexibility. Server utilization is normally very low between 10 percent and 30 percent in the one applicationper-server deployment model, which is a very inefficient use of server resources. Each server needs to be large enough to handle spikes in workload, but normally will need only a small part of the server capacity. Figure 1 illustrates this point, showing many small servers running a single application instance. Each one of these servers needs to have enough headroom to meet peak capacity requirements and cannot share headroom with other servers that need more capacity or have excess capacity. If these servers could share headroom, loaning it out or borrowing it as needed, they would have higher utilization rates. By consolidating multiple applications on a single larger server, where resources shift dynamically from application to application, the workload peaks and troughs tend to even out, and the total compute requirement is less variable. The more applications that are consolidated the more even the server usage. Applications that are consolidated on a larger server benefit from shared headroom, so consolidating applications can lead to much higher server utilization as excess capacity is reduced significantly. Figure 1. Consolidating and sharing headroom in large symmetric multiprocessing servers. 2 Improved server utilization means more efficient use of server resources, which improves ROI and reduces the total server hardware required to meet workload requirements. Consolidating many older and smaller servers onto fewer larger and newer servers provides many benefits beyond improved utilization. The newer servers will have more capacity, better performance, better energy and space efficiencies, improved availability features, and will be easier to manage. Requirements for Consolidation Servers used for consolidation must provide scalability and high capacity, high availability, and simple upgrade paths. They also must enable reuse of existing applications and have effective virtualization and resource management tools. Since applications are combined on consolidated servers, these servers need the capacity to handle dozens of workloads of all types. The performance of each application, when consolidated with other applications, must match or exceed its performance when deployed by itself on its own server. Consolidation, by definition, means putting more eggs in one basket, so a system failure will have a greater effect on application availability than if each application were deployed on its own server. Servers used for consolidation must have high-availability features, both in hardware and software, to reduce both planned and unplanned downtime. Consolidation servers must be extremely reliable so that they rarely go down. They also need to have advanced serviceability features so they can be reconfigured, upgraded, and repaired with minimal or no downtime. Consolidation servers are mainly used to run older applications in a newer environment, so they must be able to run legacy applications as well as new applications. A consolidation environment will have many workloads of different types, and these various workloads all will have specific patch, resource, security, and performance requirements. In many cases the operating system will have enough tools to manage multiple applications, but in other cases applications will require separate environments to run effectively. Virtualization and resource management tools are required so that the pool of resources in a consolidation server can be partitioned and deployed as needed for multiple applications. Virtualization enforces application separation, and resource management guarantees the performance requirements of each application are met. Consolidation on Large, Vertically Scalable SMP Servers Large symmetric multiprocessing (SMP) servers, such as Oracle s SPARC M6-32, have dozens of processors and I/O slots, and terabytes of RAM, all housed in a single cabinet that can be flexibly deployed in a single massive OS instance or separated into resource managed domains. In essence, vertically scalable servers are large pools of resources that can support dozens of workloads of various sizes and types to simplify consolidation and application deployment. New applications can be deployed on a large SMP server, eliminating the need to install a server for each new application. Existing applications can grow by taking advantage of the extra headroom available. 3 Vertically Scalable High-End SMP Servers All servers consist of the same essential components, but different server architectures combine, connect, and utilize these components in different ways. Vertically scalable servers generally larger SMP servers hosting eight or more processors have a single instance of the OS to manage multiple processors, memory subsystems, and I/O components, which are contained within a single chassis. Most vertically scalable servers, such as Oracle s SPARC M6-32 server, also can be partitioned using virtualization tools to create multiple instances of the OS using subsets of the server s resources. Virtualization tools are used to share or separate resources as required based on the workload and the security and availability requirements. In a vertically scalable design, the system interconnect is commonly implemented as a tightly coupled centerplane or backplane that provides both low latency and high bandwidth. In vertical or SMP systems, memory is shared and appears to the user as a single entity. All processors and all I/O connections have equal access to all memory, eliminating data placement concerns. Oracle s high-end SPARC SMP servers have provided linear scalability since 1993, demonstrating the value of tight, highspeed and low-latency interconnects. The cache coherent interconnect maintains information on the location of all data, regardless of its cache or memory location. There are no cluster managers or network interconnects in SMP servers because the internal interconnect handles all data movement automatically and transparently. Resources are added to the chassis by inserting system boards with additional processors, memory, and I/O subassemblies. Vertical architectures also can include clusters of large SMP servers that can be used for a single, large application. High-end SMP servers greatly simplify application deployment and consolidation. Large SMP servers have a huge pool of easily partitioned processor, memory, and I/O resources. This pool of resources can be assigned dynamically to applications using Oracle Solaris Resource Manager and manipulated using standard systems management tools like Oracle Enterprise Manager Ops Center. SPARC M5-32 and SPARC M6-32 Server Consolidation Technologies The following sections examine the consolidation technologies that enable the deployment of many applications together to improve system utilization, optimize the use of computing resources, and deliver greater ROI from IT investments. On the following page, Figure 2 shows the various levels of virtualization technologies available, at no cost, on the current SPARC Enterprise M-Series servers from Oracle. At the lower tier of the virtualization stack is the SPARC platform. The SPARC platform provides the first level of virtualization, the Dynamic Domains feature of the SPARC Enterprise M-Series (also known as physical domains or PDoms), which are electrically isolated hardware partitions, meaning they can be completely powered up or down and manipulated without affecting any other PDoms. 4 At the second level of virtualization, each PDom can be further split into Hypervisor-based Oracle VM Servers for SPARC partitions (also known as LDoms). These partitions can run their own Oracle Solaris kernel and manage their own I/O resources. It s not uncommon to have different versions of Oracle Solaris running different patch levels under Oracle VM. Oracle VM is also recognized as an Oracle hard partition for software licensing purposes. The third level of virtualization is Oracle Solaris Zones, the finest grained level of virtualization, and a feature of Oracle Solaris. Each zone in Oracle Solaris Zones shares a common Oracle Solaris kernel and patch level. They have significant advantages of flexibility when it comes to creation and reboot, and are extremely fast and lightweight. Each of these instances of Oracle Solaris can use Oracle Solaris Resource Manager to limit CPU or memory that an application can consume, usually managed with Oracle Enterprise Manager Ops Center. All of these virtualization techniques are very useful for consolidating many applications onto a single server. The next few sections describe these virtualization and resource management technologies in more detail. For the purpose of running performance-critical workloads, it is possible to configure the Oracle VM Server for SPARC domains, so that each domain is directly attached to its own PCIe slots. While this limits the total number of Oracle VM Server for SPARC domains per PDom, it provides measurable performance and isolation benefits. The Oracle SuperCluster M6-32 utilizes this type of Oracle VM Server for SPARC domains, but this type of configuration can equally be applied to both of the SPARC M5-32 and SPARC M6-32 platforms. Figure 2. Virtualization technology stack on the SPARC M5-32 or SPARC M6-32 servers. Dynamic Domains As mentioned above, Dynamic Domains (also known as physical domains or PDoms), is a feature that enables electronically isolated partitions. PDoms make it possible to isolate multiple applications and multiple copies of the Oracle Solaris OS on a single server. The Dynamic Domains feature enables 5 administrators to isolate hardware or security faults and constrain their exposure to each domain. The result is a superior level of system availability and security. The Dynamic Domains feature is now in its sixth generation, having previously been available in Oracle's SPARC Enterprise M-Series servers, making it the most mature and established partitioning option in the UNIX server market. As discussed below, PDoms can be further virtualized by running Oracle VM Server for SPARC, which allows multiple independent Oracle Solaris instances to coexist within the same physical domain. With Dynamic Domains, software and hardware errors and failures do not propagate themselves beyond the domain in which the fault occurred. Complete fault isolation between Dynamic Domains limits the effect on applications of any hardware or software errors. This helps to maintain a high level of availability in these servers, which is necessary when consolidating many applications. The Dynamic Domains feature separates the administration of each domain, so a security breach in one domain does not affect any other domain. Oracle VM Server for SPARC Oracle VM Server for SPARC provides full virtual machines that run independent instances of the operating system and are available on all of Oracle's SPARC T-Series and new SPARC M-Series based platforms. Called LDoms or Logical Domains, each operating system instance contains dedicated CPU, memory, storage, console, and cryptographic devices. LDoms are unique in the fact that many of the virtualization functions are provided natively by the underlying hardware, and that both CPU and memory are directly assigned to domains without incurring any virtualization overhead. I/O can be directly assigned either to domains, with the benefit of higher performance, or can be virtualized with the benefit of increased utilization of hardware resources and the ability to use live migration. The number of physical threads limits the number of possible domains in the system, although there is an upper limit of 128 domains per server or PDoms in the case of the SPARC M5-32 or SPARC M6-32. If the root domain model is being deployed as in Oracle SuperCluster M6-32, where domains are assigned exclusive ownership of complete PCIe slots, then the number of this type of domain is actually limited by the quantity of I/O cards available in the PDom. The use of Oracle Solaris Zones within the LDom domain creates a third layer of virtualization, mitigating the effect of this reduction in Oracle VM Server for SPARC domain count. Oracle VM Server for SPARC has the ability to perform live migration of a domain from one system to another. As the name implies, the source domain and application do not need to be halted or stopped. This allows a logical domain to be migrated to another PDom on the same server or a different server. While the use of live migration in Oracle VM Server for SPARC implementations is typical, the expected primary workloads on the SPARC M5-32 or SPARC M6-32 platforms are likely to require the most performant use of I/O, which will preclude the use of live migration. However, there may be a number of secondary workloads that could be placed on the SPARC M5-32 or SPARC M6-32 platform for which live migration would be ideal. By layering Logical Domains on top of Dynamic Domains, organizations gain the flexibility to deploy multiple operating systems simultaneously onto multiple electrically isolated domains. These domains all run Oracle Solaris, which can additionally host Oracle Solaris Zones to create yet another layer of virtualization. 6 Oracle Solaris The Oracle Solaris OS is very efficient at scheduling large numbers of application processes among all the processors in a given server or domain, and dynamically migrating processes from one processor to the next based on workload. For example, many enterprises run more than 100 instances of Oracle Database on single SPARC servers using no virtualization tools. Oracle Solaris is able to effectively manage and schedule all the database processes across all the SPARC cores and threads. With this approach, a large vertically scalable server can assign resources as needed to the many users and application instances that reside on the server. Using the Oracle Solaris OS to balance workloads can reduce the processing resource requirements, resulting in fewer processors, smaller memory, and lower acquisition costs. Oracle Solaris increases flexibility, isolates workload processing, and improves the potential for maximum server utilization. Oracle Solaris Zones In a consolidated environment, it is sometimes necessary to maintain the ability to manage each application independently. Some applications may have strict security requirements or might not coexist well with other applications, so organizations need the capability to control IT resource utilization, isolate applications from each other, and efficiently manage multiple applications on the same server. Oracle Solaris Zones technology (formerly called Oracle Solaris Containers), available on all servers running Oracle Solaris, is a software-based approach that provides virtualization of compute resources by enabling the creation of multiple secure, fault-isolated partitions (or zones) within a single Oracle Solaris OS instance. By running multiple zones, it is possible for many different applications to coexist in a single OS instance. The zones environment also includes enhanced resource usage accounting. This highly granular and extensive resource tracking capability can support the advanced client billing models required in some consolidation environments. Oracle Solaris Resource Manager Oracle Solaris Resource Manager is a group of techniques that allow the consumption of CPU, memory, and I/O resources to be allocated and shared among applications within an Oracle Solaris instance including Oracle Solaris Zones. Oracle Solaris Resource Manager uses resource pools to control system resources. Each resource pool may contain a collection of resources, known as resource sets, which may include processors, physical memory, or swap space. Resources can be dynamically moved between resource pools as needed. Also, with Oracle
Recommended
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x