As companies try to consolidate their multiple low-end servers onto larger machines, there are vast technical pitfalls that can make their quest for lower operating costs unrealistic. In particular, the degree of support offered by different operating systems varies markedly between the consolidation platform options.

The base platform options are fairly clear: Windows NT and its more robust update, Windows 2000, Unix (in all its flavours) and mainframe operating systems. But how well do these environments support the operation of multiple systems on the same engine, and what specialist capabilities do high-end environments require to successfully support consolidation?

	Alan Priestley, Intel: “Moving operating systems is always a big area of concern.”

Above all, there are several key software technologies that are required for managing multiple, consolidated applications: partitioning (for segmenting the applications), workload/resource management (to ensure individual applications receive a fair share of systems resources such as processor cycles, memory, I/O, and so on), load balancing (to distribute application demands), and failover clustering (to ensure the applications using one CPU are swapped to another in the case of a failure). However, those only start to appear at the higher end of the server market.

Least pain
For many companies with departmental servers, the operating environment today is Microsoft Windows NT. On the surface, consolidation on a larger NT server might seem the least painful option, but, says Andy Butler, an analyst at Gartner, NT consolidation is a long way behind the Unix world. "Consolidation is well understood in Unix and there are sophisticated middleware products to support it. In the NT space, the idea of consolidation is really very new – it is only in the last six to 12 months that we have seen NT users get interested in consolidation," says Butler.

Ian Benn, marketing director at systems and services company Unisys, says consolidating services onto NT is not a good idea anyway – far better to consider a move to Windows 2000. "If you're going to consolidate, you're going to need an operating system that's not going to crash. I wouldn't want to do server consolidation under NT." In terms of robustness, "there aren't any real problems or significant issues with Windows 2000."

But only the higher versions of Windows 2000 are likely to be applicable to consolidation initiatives. Windows 2000 Server supports four-way servers; Advanced Server supports eight-ways, failover clustering and network load balancing; and Datacenter Server can cope with 32-way boxes and has, in addition, a process control manager.

Advanced Server will be the natural choice for the small business that wants to consolidate a few NT or 2000 servers into a single machine. But the extra capabilities of the product and the mainframe reliability approach taken by Microsoft and its partners with Datacenter Server will give it added appeal to the large enterprise.

"Datacenter has the same code as Advanced Server, but the third-party program certification is completely different," says Benn. "There's no problem with a program that runs in user mode, but the big catch is if it touches the kernel, it will be stopped unless it has been certified. Datacenter Server itself is only certified on a few specific vendors' hardware and any set up is given weeks of testing under stress loading before going live."

Ring-fenced apps
While operating system stability is important, so too is application stability. Consolidating several applications onto one server can cause instability in individual applications. Says Andy Jordan, Unisys consolidation manager "In running a mixed workload of applications on one instance of the operating system, you are breaking an unwritten rule of Windows, which is that you only run one big application on each server."

The ability to ‘partition' a server so that it can run several operating systems at the same time and appear as several different servers becomes particularly important in consolidating mission critical applications. Each application can have an operating system of its own in its partition, so even if the operating system or the application crashes, everything else continues unaffected. IBM, for example, is applying the historical strengths of its VM operating system to offer organisations the ability to run hundreds of different virtual Linux or NT environments on its mainframes, all kept isolated by VM.

Analysts are impressed by the approach. "[A] compelling reason for deploying a mainframe for server consolidation is to take advantage of the platform's system resource management capabilities. For example, the [IBM] z900 is capable of automatically and continuously adjusting the allocation of system resources among users and applications based on established business priorities," says Mary Hubley, an analyst at Gartner. "End user service levels can be defined in terms of acceptable system response times to user requests for service, relative user priorities, completion times or other parameters."

That is a point echoed by analysts at consultancy DH Brown. "Hard partitioning meets many of the requirements for consolidating applications onto a single server. Each application can run in its own partition, effectively recreating the conditions under which they were originally deployed. [Many] implementations of Unix partitioning schemes, however, involve trade-offs that make Unix partitions unwieldy for that purpose." Some require individual management of partitions, some consume overhead when managing the partition. On the other hand, Unix systems have for long offered resource management for allocating systems resources to different applications.

Another, less technical advantage to partitioning is the placation of software vendors. For performance reasons and to avoid conflict for resources, package vendors recommend that their enterprise application sits alone on a server. Inevitably, there is a tendency to blame any faults on any other application sharing the same environment. Partitioning removes that area of friction.

While Windows 2000 Datacenter Server's hardware partitioning is as yet unproven, that aspect of technology is well established on Unix and mainframe platforms. "It's difficult to conceive of Intel with Microsoft covering all environments from one end to another," argues Colin Grocock, new business director, IBM eServer. He claims IBM's zSeries mainframes can run thousands of separate instances of Linux, for example. This degree of partitioning is extreme – and, as Butler points out, "very few organisations are going out and buying a new mainframe to do consolidation". But many of IBM's other offerings are capable of partitioning – albeit to a lesser extent – as are Sun Microsystems' Solaris-based servers. For instance, the IBM e690, known as ‘Regatta', is the first of IBM's AIX Unix servers to offer the facility.

But it is not the first implementation. Partitioning came to prominence in the Unix world in 1997 with the release of the Dynamic Domains for its E10000 server. "Sun heralded Dynamic Domains as Unix's nearest equivalent to the logical partitions familiar to mainframe users," say analysts at DH Brown.

Shareware
Sharing multiple operating environments on one server remains risky, says Butler at Gartner. "Abbey National has done this with the Unisys ES7000 to run both NT and Unix. As far as I am concerned, the Unisys box is the only server on the planet that can really let you do this. But sharing operating systems on the same box isn't something you wake up one morning wanting to do."

In fact, Unisys' CMP (Cellular Multi-Processing) technology allows Windows 2000 and Unix to reside on the same system as the Unisys 2200 operating system, a big selling point for longstanding Unisys customers looking at consolidation.

But for those with mixed environments who wish to consolidate, switching applications to a different operating system may be equally difficult. While most operating systems have programs that allow them to run software designed for other processor architectures and operating systems, the performance hit is usually so great that the only realistic option is to ensure applications that are present on one set-up are available on the destination set-up or can be easily ported.

"Moving operating systems is always a big area of concern," says Alan Priestley, strategic marketing manager at Intel. "It's hard to prove that what you've got is going to operate successfully on the new system. Some environments are completely different. You might have to change the management environment. With bigger systems, there may be a change in processor architecture which means a recompilation of code."

Mark Byatt, marketing director at systems integrator Morse, says a change of operating system can reveal mistakes made in previous years. "Inevitably, when you consolidate, you will be retiring old structures that will not fit into the new scheme. With consolidation you swiftly learn which of the decisions you made two, three or five years ago were the right ones."

As companies look to cut their overheads by reducing the number of servers in their organisations, so they have discovered that what was good for the low end just does not cut it at the high end. The decision they have had to make is whether the benefits of consolidation outweigh the risks and the costs (see article, Cost concentration). While Windows 2000 Datacenter Server may not have all the capabilities the best Unix and mainframe operating systems have to offer, its natural fit with existing set-ups means that it will still find favour among many organisations as a consolidation platform. But for the largest enterprises, Unix and mainframes are still the starting points for consolidation.