Exploring Advanced Features

There's a range of advanced capabilities that each vendor tries to use to differentiate themselves . In the following sections, you'll look at some of these advanced features for the key platforms.

Sun SPARC Platform Features

In the preceding Sun SPARC section, I touched on some of the advanced features of the Sun platform. You'll now look at these features in more detail and see how they apply to a WebSphere environment.

Dynamic System Domains

Dynamic System Domains (DSDs) are a Sun technology found in its SunFire family of servers and the enterprise E10K server. Basically, domains provide the ability to split a physical frame (chassis) into separated servers. Each server is essentially its own machine, with dedicated CPU and memory, PCI interface cards and buses, and so forth.

There's a common component shared across multiple frames of the system ”this component, known as the Fireplane , is the overall system interconnection bus and operates in the fashion of a switch.

Domains allow you to effectively run an entire platform in a single chassis. If your requirements state that you need processing power across your tiers of up to 24 Sun UltraSPARC CPUs, then through the use of domains and an F6800, you can partition your frame into several servers.

The domains can be arranged within the chassis so that if you were to operate two WebSphere application severs as part of your domain configuration within a particular F6800 frame, the two WebSphere application server domains could be split so that they wouldn't share any common components . That is, it's possible to design the layout so that each common or clustered application server (which is a domain) would operate on differential power supplies , memory and CPU boards , system controllers, and I/O controllers. Further, you can configure the system so that the Fireplane interconnect is segmented down the spine of the frame.

When you partition domains in this manner, you're effectively building two independent servers (for two domains) or a pair of servers (for four domains).

The other benefit of the domains is that most components within the SunFire family of servers are hot-swappable and hot-pluggable. If your data center operates a capacity-on-demand model or you use computer virtualization, SunFire servers and domains allow you to add and remove capacity and components in real time, without any customer or operational impact.

Hot-swappable components include the following:

• CPUs and memory boards

• Power supplies

• The Fireplane interconnect switch (hot-pluggable only)

• I/O assemblies

• PCI adaptors

This type of configuration can allow you to operate a configuration such as a dual-node database cluster and a dual-node WebSphere application server cluster within a single chassis.

Although the entry costs of this sized frame are higher for lower-end configurations, the cost benefit starts to be realized when you take into account the power and flexibility associated with the advanced domain and SunFire technologies. Facilities management costs can be greatly reduced with this type of configuration, and should a multisite option be a consideration, you could opt for a dual, lesser- powered F6800 configuration split over two sites or a dual F4800 configuration, which supports up to 12 CPUs and 96GB of memory.

Grid Computing

Grid computing is starting to become more apparent in the finance and scientific/engineering markets. Although it can't be implemented via WebSphere (without some serious development effort), WebSphere environments can interface with a grid-based computing environment.

Essentially, grid computing is a form of parallel clustering. This type of clustering was made famous by the cluster farms associated with the graphical rendering of the film Titanic , whereby clusters of hundreds of servers are connected together through a certain topology and with some specific application coding and server software, allowing all servers in the cluster to operate as one.

It's not possible to use WebSphere in this fashion ”WebSphere uses its own clustering technologies (which I'll continue to focus on). However, if you're involved in a site that requires vast amounts of computing power, you'll need to interface with a grid-based environment.

Note  

Grid computing isn't necessarily a Sun concept. Many commercial vendors , including IBM and HP, have their own versions of grid computing. Various standards are starting to emerge that allow grid-computing environments to share resources across multiple vendor systems. Furthermore, Linux has a number of open-source implementations that are along the same lines as grid computing. The most common one is the Beowulf cluster ”an open -source parallel computing framework giving the ability to interconnect vast numbers of commodity Linux servers together to operate in a parallel cluster environment.

IBM PowerPC Platform Features (pSeries)

The PowerPC/Power range of IBM servers exhibited a number of technologies that provided high availability and other advanced computing options. You'll look at two that are similar to those you've explored for the Sun platform range.

Logical Partitioning (LPAR) and Dynamic Logical Partitioning (DLPAR)

IBM's system partitioning technology is called DLPAR. LPAR, on the other hand, is the predecessor of DLPAR and is a fixed version of DLPAR. Essentially, LPAR allows an IBM pSeries server to be partitioned. This partitioning, like SunFire domains, allows you to distribute CPUs, memory, and I/O adapters across multiple logical servers within a single chassis. DLPAR is an advancement of LDAR and allows the components within the various logically portioned servers to be dynamically allocated (or deallocated) to and from partitions while they're active.

This extends the notion of computing on demand to a new level by allowing unused resources to be allocated to a production environment during peak times and deallocated during off-peak times for testing or other purposes.

Another popular use of this technology is to house a spare set or pool of resources (in other words, CPU, memory, and I/O adaptors such as network interface cards and so on), which are available to any of the other live partitions in the frame. In the event of component failure or adverse peak loads, DLPAR can let you absorb unforeseen peaks and problems with the performance of your WebSphere applications.

The inner workings of LPAR and DLPAR are beyond the scope of this book; however, I'll discuss DLPAR again throughout later chapters when you delve into AIX operating system tuning during Chapter 5.