The Server Consolidation Era

For many businesses, the proliferation of Windows NT servers in their enterprise, first as file and print servers, then as application servers, has created a complex and difficult-to-manage computing environment. The lack of trained Windows 2000 (Win2K) and NT IT staff adds to this complexity. When you add the constant stream of server upgrades occurring in the industry, you create a situation in which consolidating onto larger servers becomes favorable for companies wanting to save money. This consolidation will be the mantra of many hardware companies selling Win2K solutions in the next year or two. (For more information about server consolidation, see Sean Daily, "Making the Case for Server Consolidation," page 95.)

The capabilities of Windows 2000 Advanced Server (Win2K AS) and Windows 2000 Datacenter Server (Datacenter) also offer companies the opportunity to perform server consolidation. For server consolidation, administrators can add larger SMP systems such as 8-way Profusion servers and 16-way and 32-way systems (which are just around the corner). Administrators can also put more small servers into the same server chassis—an approach that works well when running Windows 2000 Server Terminal Services and Microsoft Internet Information Services (IIS). Given recent and expected advancements in dynamic load balancing and partitioning of application workloads, Windows server consolidation makes sense.

Nearly every Win2K hardware vendor has a total cost of ownership (TCO) model that justifies the replacement of large numbers of servers with a smaller number of powerful SMP boxes or dense rack-mounted systems. Simplifying system complexity, reducing IT staff, and improving reliability results in measurable cost savings. If your company runs Microsoft Exchange Server on eight servers, running it on two quad servers might be cheaper and easier. In the past, software vendors have encountered resistance to their NT products from IT staff (particularly in companies using mainframes and UNIX) because of the number of servers necessary to deploy the applications on NT. Administrators also mention server size (i.e., the footprint) as a concern, but in the next year the footprint for Win2K servers is likely to become smaller.

Servers are a rapidly depreciating asset, and buying a leading-edge and powerful server that has the longest life span makes sense. However, a more cost-effective and flexible approach for organizations is to adopt rack-mounted systems to which you can add servers as blades (i.e., computers built on one plug-in board) to a backframe (i.e., a chassis with slots for blades, connectors, and I/O ports). As chip technologies advance, you can replace the server blades. Therefore, companies such as IBM, Compaq, Dell, and Data General have a great interest in developing the smallest possible (thinnest) rack-mountable servers. An example of a thin-profile server is Dell's PowerEdge 6350. This rack-mountable server sells for $7694, and Dell is positioning it for corporate data centers to run database and enterprise resource planning (ERP) applications.

Thin-profile servers are becoming key components in some major corporate deployments. In January 2000, Akamai Technologies named IBM as its key hardware partner, bypassing Dell, Compaq, and VA Linux Systems (the Linux hardware specialist) for the contract. Akamai sells Web-caching software. The reason Akamai cited for selecting IBM is the company's expertise in building thin Linux servers for the major ISPs. IBM's 1U (1.75") server is the industry's thinnest and densest rack-mountable server. In April 2000, IBM plans to announce a new set of thin servers: the 1U server code-named Intimidor, a 3U (5.25") 2-way server code-named Illinois, and a 4U (7") server code-named Assault.

We're seeing the introduction of systems that use load-balancing applets to create enterprise-class applications based on Redundant Array of Inexpensive Computers (RAIC). Although RAIC lags the development of SMP, it offers attractive sizing options for companies growing to enterprise-class sizes. The sizing options might also make rack-mountable systems more popular than SMP servers.

If your organization is storage-centric, then it might recognize storage to be a long-lived asset. You can design consolidation so that you attach server blades directly to either network-attached storage (NAS) devices or to Storage Area Network (SAN) implementations. In the future, large enterprises will probably con-sider storage first when deploying enterprise solutions.

In the past, the low acquisition cost of NT servers tended to fragment IT assets deployed on those servers, depending on the department and site location. One central design goal of Win2K has been to provide an overall systems management capability that makes taking inventory, monitoring status, and controlling servers easy to perform from a central location. The integration of the Microsoft Management Console (MMC) and Active Directory's (AD's) data makes this central management possible. Not only can IT staff manage the NT infrastructure, file systems, clusters, and applications centrally, but they can also use organizational units (OUs) to delegate departmental resources and security functions. NT server proliferation has been a barrier to NT applications' entry into the enterprise. But for many of the reasons just mentioned, this proliferation is about to change.