Hardware Support
Paradoxically, while hardware costs fall almost daily, the requirements for scalability and availability that customers impose on data centers simultaneously increase. Many systems now come with the ability to handle multiple CPUs and advanced storage facilities such as RAID arrays. To make the best use of this hardware, the OS must support the equipment at a system level, rather than relying on additional applications and device drivers.
Hardware and software purchase price is only a part of the cost of ownership. Reducing the total cost of ownership requires using hardware and software that is not only low in terms of its initial cost, but also in its ongoing support costs. In particular, the modern data center requires the highest levels of availability possible along with exceptional management capabilities. Hardware needs to be resilient, but the operating system must be able to take advantage of hardware capabilities, both for monitoring the situation and to when providing a reliability solution.
An operating system that can monitor the entire installation, detect anomalies, and then provide an intelligent solution provides the best overall reliability. Support needs to be provided both on a single system basis, such as when a UPS switches to battery rather than on-line power, and also within a network application environment. A data center, for example, must be able to handle application server component failure, either through the use of a network load balancing technology or solutions such as SMP and clustering so that the effects of the failure are not passed on to the end users.
Providing this level of availability goes beyond the facilities found in standard hardware technologies. In addition, you also need a support network standing behind the hardware and guarantees that the operating system is compatible with, and can take advantage of the hardware features. Any additional hardware and software used within the final solution configuration should also be subject to the same testing and standards as the base platform.
CPU, SMP and Clustering Support
Both Solaris 8 and the Windows 2000 Server product family support multiple-CPU configurations. Solaris is supported on both SPARC and Intel hardware. The SPARC microprocessor architecture is a proprietary Sun design. It is only used in desktop and server solutions from Sun and two other low volume manufacturers. This means that Sun must amortize the huge cost of design and test across a relatively small number of systems. This contrasts with Intel, whose designs are also proprietary, but which can spread the cost of design across millions of computers sold by hundreds of manufacturers.
The fourth generation of SPARC processor is the UltraSPARC II, now standard on all systems. Solaris 8 supports SMP solutions on both SPARC and Intel solutions. The fifth generation, UltraSPARC III was just announced as being available in a few low-end systems. Solaris 8 is required for the new UltraSPARC III systems.
Within currently available SPARC and Intel platforms it is possible to configure 64 CPUs within a single system. For data center applications, up to four nodes can be clustered together to provide the maximum performance or fault resilience. The clustering technology requires additional hardware to operate, including a proprietary 1 GB Scalable Coherent Interconnect (SCI), which increases the overall cost of the clustering process.
The distribution of the workload is automatically handled by the operating system to make best use of the available power, or it can be controlled using the Sun Cluster Manager and the Solaris Resource Manager to allot individual tasks to each processor within a single node or cluster.
All versions of Windows 2000 support the Intel Pentium processor family and will support the forthcoming Intel Itanium and other IA-64, 64-bit Intel processors when they become available. Windows 2000 Server supports 4-way SMP systems, and Advanced Server extends this support to 8 CPUs and adds clustering support for up to two nodes. The clustering technology is built into Windows 2000, instead of being a separate component.
Windows 2000 Datacenter Server extends the multiprocessor support to up to 32-way SMP systems and clustering for up to 4 nodes (128 CPUs). The Datacenter Server supports two different types of scalability at the hardware level, scale-out and scale-up. The scale-out model has network computing as its design center and follows the traditional Windows path. Capacity can be increased by adding more servers, which Datacenter Server supports using a combination of clustering and Network Load Balancing. The scale-up model follows the traditional UNIX path. You add more processors, RAM and storage to an existing server. This greatly increases the level of investment in a single box, which is, effectively, a single point of failure.
Both the Windows 2000 Advanced Server and Datacenter Server products take advantage of Physical Address Extension (PAE) support, which is part of the Enterprise Memory Architecture (EMA). PAE enables 32-bit Intel based systems to address more than the normal 4GB memory limit. On Advanced Server, this enables designs of up to 8GB of addressable memory. On the Datacenter Server product this ceiling is increased to 64GB. The benefit is that systems employing PAE exhibit very little paging, thereby increasing performance while still being supported under the cost-effective 32-bit processor model.
In addition to the SMP and clustering technology, Windows 2000 Datacenter Server supports hardware-partitioning, which allows you to run multiple instances of the operating system on the same physical hardware. This level of control was previously available only on the larger Unix systems, such as Sun’s Starfire server.
Unlike Solaris, Windows 2000 clustering can be operated over the inexpensive SCSI bus standard. Additional SCSI cards can provide resilience for the cluster communication without the need for additional software.
For resource management, Windows 2000 includes a tool for allocating individual resources for each application. On Windows 2000 Server and Windows 2000 Advanced Server, this facility is handled by the Application Programming Interface (API) to the operating system, allowing software vendors to maximize the performance of their applications. On Windows 2000 Datacenter Server, this control is achieved by using the Process Control tool, which can allocate individual resources, including processors and memory, to each application. A summary of the CPU, SMP, and Clustering support is shown in Table 1 below.
Table 1: CPU/SMP Support
|
Feature
|
Solaris 8
|
Windows 2000 Server
|
Windows 2000 AS
|
Windows 2000 DS
|
Platform
|
SPARC
|
Intel
|
Intel
|
Intel
|
Intel
|
Processor Bit Width
|
32
|
64
|
32
|
32
|
32
|
32
|
Maximum CPUs per node
|
64
|
64
|
64
|
4
|
8
|
32
|
Maximum Nodes per cluster
|
4
|
4
|
0
|
2
|
4
|
Advanced Job Control
|
Y
|
Y
|
Y
|
API
|
API
|
Y
|
Maximum Addressable Physical Memory (GB)
|
4
|
64
|
4
|
4
|
8
(PAE)
|
64 (PAE)
|
|
