Download
Please get the complete article here. It also provides you with a detailed installation example using the core components of the "Adaptive Oracle Grid". That is Solaris 10 Container, Oracle 11g Clusterware, Oracle 11g ASM (Automatic Storage Management) and Oracle 11g Database.

The "Adaptive Oracle Grid"

Grid technology allows disparate systems to be pooled and managed as a common computing resource. Applications in the grid are being served by many computers in parallel at the same time. There's no question that grid computing can deliver significant cost savings and boost the quality and flexibility of your IT services. The question is: Do you have the right infrastructure in place to take full advantage of grid computing?

"Scale out" with Oracle Grid

Oracle Real Application Clusters (RAC) is Oracle's core technology to offer grid computing in the database tier. RAC enables a database to run in parallel on distributed servers. Based on this architecture, Oracle's vision is that customers can quickly and easily create a large-scale computing infrastructure from small components like server blades. This Grid architecture provides the ability to "scale out" by adding new servers as additional capacity is needed.

But customers going the "scale out" approach will be faced with a couple of new challenges:

Increasing amount of servers to integrate into the datacenter, to monitor and to maintain.
Increasing amount of Operating Systems to install and to maintain.
Free capacity (i.e. CPU and Memory) is fragmented into small pieces across many small servers and across different RAC systems. Thus it is difficult to have the free capacity where it is really needed at any point in time.
"Scale out" means to add new servers to a cluster, as needed. But how long does it take in the datacenter reality to order the new server, integrate it into the infrastructure, install the OS, install Oracle software and finally configure as new cluster node ??

Figure 1: The Server Sprawl

Sun customers love "scale up"

In the early 1990s Sun Microsystems began to leverage large-scale symmetric multiprocessing servers. Today these systems provide mainframe-class features and data center reliability for the open systems market. And Sun customers have come to appreciate the advantages of high-end servers.

In interaction with the Solaris 10 OS, Sun Enterprise Servers deliver an almost linear scalability up to 256 CPU-cores (512 parallel threads) for every type of application. Thus many application services can be consolidated on just one single Sun Enterprise Server, and the servers spare capacity is available as one big pool to all services. Even better, additional server capacity - if needed - can be added online (without a service interruption), without the installation of a new OS and without the need to reconfigure any software. The applications running on the server will immediately benefit from dynamically added resources (CPU and memory).

Make the most of it

Wouldn't it be great to have the advantages of "scale out" and "scale in" both at the same time ? And here it is: The Adaptive Oracle Grid !

Figure 2: The architecture of the "Adaptive Oracle Grid"

Oracle RAC is implemented in Solaris 10 Container (aka Zones). Each Container can have a complete own and separate Oracle software stack containing Oracle Clusterware, Oracle ASM and Oracle Database. Several Oracle-RAC-Zones can be consolidated on a small number of Sun Enterprise high-end Servers. Thus providing the options to "scale-up" and "scale-out".

Each Container in the Grid can be re-sized online and dynamically by the Solaris Resource Management. The administrator can assign CPU-shares to a Container. This is a "minimum-guarantee"-model: A specified fraction of systems total capacity (number of CPU-shares) is guaranteed for the Container. Furthermore the Container will get more CPU, if unused resources on the server are available. But if the Container does not require it's CPU resources at any point in time, these free resources will be available to all of the other Container. This self-balancing dynamical management provides the best server utilization and thereby is the most effectively.

Solaris DISM (Dynamic Intimate Shared Memory) provides shared memory that is dynamically re-sizable. This enables Oracle Database 11g AMM (Automatic Memory Management) and allows to manage the memory online and automatically by just one single Oracle parameter for the complete instance (memory_target).

It is important to understand, that Solaris Container are not additional Solaris installations on physical or virtual servers. Container provide an environment that appears to be a complete, standalone Solaris 10 OS installation. But actually it is just a kind of an additional running instance from the existing Solaris OS on the server. In this way, Container provide virtual Solaris Operating Systems - in contrast to hypervisor-based technologies that provide virtual machines (or virtual servers).

And the difference to virtual maschines is incredible: New Container are implemented in minutes (because there is no new OS installation), without any performance degradation (because there is no hypervisor) and without any 3rd party software or licenses (because everything is included in Solaris).

Even large scenarios as shown in figure 2 - having five RAC systems with 40 cluster nodes - can be made of just four Solaris OS installations. That means only four Operating Systems to install, to monitor and to maintain. What a formidable difference !!

Figure 3: Solaris Container provide virtual OS

Advantage for the "Adaptive Oracle Grid"

Explore the advantages of the "Adaptive Oracle Grid".

Usability: Every free CPU second and every free memory-Bit is available to any RAC system in the "Adaptive Oracle Grid". At any point in time.
Efficiency: Free spare capacity is not fragmented in small units across multiple servers. Instead it is pooled in larger blocks on just few serves and will be used effectively to support peak utilizations.
Self-balancing: Provide each container with a guaranteed amount of CPU resources (CPU-shares). But at the same time don't block any additional demands, if free server capacity is available. That is a win-win situation for every container.
Scalability: Nearly unlimited scalability with the "scale-up" of Sun Enterprise high-end servers and the "scale-out" of Oracle RAC.
Speed-up: As Solaris Container are not new OS installations, you can provide several new (virtual) cluster nodes just in few minutes.
Managebility: No sprawl of servers and OS installation. Even largest landscapes can be made of few servers and the same few amount of Solaris installations.
Online maintenance: Benefit from the high-end features of Sun Enterprise Servers. Maintain CPU, memory and IO while the system continues to operate.
Live upgrade: Upgrade and maintain (patch) the Solaris OS online using "Solaris Live Upgrade" and reduce the planned downtime to just one server reboot(1).
License cap: Solaris Container are recognized by Oracle as license cap(2). So you don't have to pay for more CPU than needed in the Container.

(1): For details please see the Sun Whitepaper "Upgrade- und Patchmanagement in einer virtuellen Betriebsumgebung" (http://de.sun.com/servicessolutions/virtualization/ressourcen.jsp).
(2): For details please see the Oracle statement to partitioning (http://www.oracle.com/corporate/pricing/partitioning.pdf).

Looking for ECO efficiency and incredible savings ?

Scale more with less. Increase the eco-efficiency of your datacenter without sacrificing performance with the Sun SPARC Enterprise T5140 server, which offers the highest compute density available: 2 CPU's with a total of 128 parallel threads in only one rack unit! This server provides world record performance throughput.

Charles Phillips, President Oracle says: "With these new Sun systems, Sun is [...] creating a new market opportunity with a high-performance, low-cost deployment alternative for Oracle grid."

You may also like to save license fees? So let's have a look at the Oracle Database 11g Standard Edition. The Standard Edition is an affordable, full-featured database. But it is is at roundabout a third of the costs of the Enterprise Edition and it already includes Oracle RAC at no additional costs. It is limited for servers (or clusters) up to four sockets.

The gist of the matter is bring both pieces together and deploy Oracle RAC on a two node T5140 cluster, that is a four socket cluster configuration. This allows you to benefit from the outstanding scalability and efficiency of these servers and at the same time using the Oracle Database 11g Standard Edition. This - as already mentioned - saves significant license fees.

Figure 4: The Adaptive Oracle Grid, optimized for TCO

Let's compare this architecture with competitive systems. IBM the entitles the POWER6 processor as "the world's fastest chip". The smallest system that is capable to hold two POWER6+ processors at their highest clock-rate of 5.0 GHz is currently the IBM Power 550. Comparing the Sun T5140 and the IBM Power 550 (both with 64 GB and two sockets) shows that:

the Sun T5140 provides much more throughput (80% more with SAP SD bechmark, 25% more SPECint_rate_base2006)
the Sun T5140 has a 35% lower TCA (Total Cost of Acquisition. Sun: 37.000 €, IBM: 58.800 €).
the Sun T5140 has a 38% lower TCO (Total Cost of Ownership. Sun: 18.500 € p.a., IBM: 29.900 € p.a.).

Figure 5: TCO of IBM Power 550 compared to Sun T5140

References
SAP SD standard SAP ERP 6.0 (2005) application benchmark, Certification Number 2008021. 2 x UltraSPARC T2+ processors provide 20.900 SAPs.
SAP SD standard SAP ERP 6.0 (2005) application benchmark, Certification Number 2008019. 32 x POWER6 5.0 GHz processors provide 177.950 SAPs. Thus approximately 12.000 SAPs for two processors.
SPEC CINT2006 Rate, 2 x UltraSPARC T2+ processors provide SPECint_rate_base2006 of 142. (http://www.spec.org/cpu2006/results/res2008q2/cpu2006-20080407-04060.html)
SPEC CINT2006 Rate, 4 x POWER6+ 5.0 GHz processors provide SPECint_rate_base2006 of 215. (http://www.spec.org/cpu2006/results/res2008q2/cpu2006-20080407-04060.html). Thus approximately a rate of 113 for two processors.
Prices according the the price lists and web shops of the vendors (as of may 2009). Support and service fees according Ideas Competitive Profiles (http://as.ideascp.com/).

Oracle Support
Please note, that as of today (March 2009) Oracle RAC in Solaris 10 Container is not yet supported by Oracle. But it is scheduled and expected for summer 2009.

Adaptive Oracle Grid