November 18, 2002 (Computerworld) --
The current generation of high-end Unix servers can be configured to support large, monolithic applications or the consolidation of many small ones. These machines—which typically support at least 32 processors—may serve as one large symmetrical multiprocessor system or as a collection of smaller individual hardware and software partitions, each serving a different application.
Advances in microprocessor design, interconnect technology, clustering, and resource and memory management have all contributed to the development of systems that offer near-mainframe horsepower and reliability, says Gordon Haff, an analyst at Illuminata Inc. in Nashua, N.H.
All high-end Unix servers feature modular designs. They usually include a collection of RISC-based quad-processor modules, each with a limited amount of local memory—and sometimes local I/O as well.
Most of the raw performance comes from taking many nodes, throwing in large amounts of memory and I/O support, and linking them with superfast interconnect technology and resource management software, say analysts.
Top-of-the-line models may support up to 128 processors and as much as 576GB of memory in a single box. Such scalability is particularly useful for tasks such as running very large Oracle databases, says Brad Carlile, a director at Sun Microsystems Inc.
Also contributing to the raw performance of such servers are superfast microprocessors, such as IBM's 1.3-GHz dual-core Power4 RISC chips, or the sophisticated interconnect technology that helps Hewlett-Packard Co.'s Superdome server deliver 64GB of I/O bandwidth.
Meanwhile, growing support for technologies such as clustering, error-correcting memory, and redundant and hot-swappable componentry have made high-end Unix servers far more reliable, says Steve Josselyn, an analyst at IDC in Framingham, Mass.
Together, such capabilities make for pretty formidable servers, says Bob Moore, IT director at Paetec Communications Inc., a local exchange carrier in Fairport, N.Y.
"If you look at it, a lot of the features and functions found on these systems are the same as those found on mainframes—but without any of the overhead," Moore says.
Physical Partitions
All of the system vendors offer partitioning technologies that let users run multiple applications on the same box, just as mainframes have done for years.
Both Sun and Tokyo-based Fujitsu Ltd., for instance, allow users to carve up their boxes into 18 separate hardware-bound partitions, each of which can run a copy of the Solaris operating system and its own applications.
Each four-processor partition is completely isolated from the others from an electrical, hardware and software standpoint so that even if one partition crashes, the other partitions will continue.
Software-based partitioning technologies, available from both IBM and HP, build on this capability by allowing users to carve servers into even smaller pieces than hardware partitioning permits, Josselyn says. While software partitions aren't as isolated and secure as hardware-based ones, they offer more configuration flexibility.
HP's Superdome, for instance, lets users create multiple soft partitions within a hardware partition. IBM's technology lets users of its p690 servers create virtual partitions as small as one processor and 256MB of RAM. Software-based partitioning schemes allow for better granularity and resource control than do hardware partitions, says Ravi Arimilli, chief architect of IBM's pSeries servers.
Layered on top of this is a growing stack of rapidly maturing resource-management software, such as Sun's Resource Manager and HP's Workload Manager, that also provide partition and systems management capabilities.
All vendors also offer features that allow capacity upgrades on demand. Users buy servers with excess processors upfront and then switch them on when needed. HP's Superdome, for example, even lets users turn off the additional resources when they're not needed.
Slicing Up a Unix Server
1. A high-end Unix server can be configured as a monolithic system for running a resource-intensive application such as a very large Oracle database. Today's largest Unix servers can support up to 128 processors and 576GB of memory.
Today's large Unix servers can be run either as big, monolithic systems . . .
2. Alternatively, the systems can consolidate multiple applications and run them in separate hardware or software partitions on the server. Each node in a hardware partition is typically a four- or eight-processor module that includes memory and sometimes local I/O. Soft partitions can be smaller. Each partition runs its own operating system image.
. . . As multiple smaller servers separated by hardware or software partitions . . .
3. Adding software partitions permits more granular allocation of resources, allowing each processor on a quad-processor node to run a different application, for example. But physical partitions offer higher levels of fault isolation and security.
. . . Or with software partitions running inside hardware partitions.
See results from our survey of more than 5,000 IT pros, and use our Smart Salary Tool to compare your pay with IT workers in similar jobs across the U.S.
See your link here
or
Other Servers Stories
See results from our survey of more than 5,000 IT pros, and use our Smart Salary Tool to compare your pay with IT workers in similar jobs across the U.S.
After weathering layoffs or pay cuts, your IT staffers may need some help getting motivated. Try these strategies for employee renewal.
No Windows geek or PC support pro should be without these must-have utilities -- and they're all free.
Get the latest news, features, opinions and more on key technology issues.
Get the latest news, reviews and more about Microsoft's newest desktop operating system.
General Mills, Genentech, San Diego Gas & Electric, University of Pennsylvania and Monsanto top the list.
Computerworld
