As an IT service provider, SI recognizes that any money it can save in its own IT operations gets directly passed to the company’s bottom line as increased profitability. So one of the most important questions SI’s executive board (which consists of six IT and business executives) needed to address from the outset of this company’s formation was, “How can we improve overall information systems efficiency, while reducing waste?”
Some answers were obvious. Systems, storage, and network management can be labor-intensive and expensive. Further, management of distributed information systems can be extremely complex (finding and exploiting unused resources in a distributed environment can be a real challenge, as can securing all the access points in distributed systems architectures. SI’s executive board recognized that the company should automate the management of systems resources whenever possible to contain management costs.
Other opportunities to improve enterprise computing efficiency were, however, less obvious. For instance, is it more energy efficient to deploy hundreds of smaller servers, or dozens of large servers? By consolidating many servers into fewer servers, SI realized that large servers, running at higher usage rates, actually burn tremendously less energy than smaller servers that run at 10 percent or so of capacity. (As proof of this concept, IBM recently announced its mainframe “gas gauge,” a measurement tool that has been used to show that certain mainframe configurations can process the workload of 250 x86-based Linux servers using only 10 to 12 percent of the energy used by these Linux servers.)
SI’s board decided to centralize management functions across fewer systems (as opposed to managing multiple, distributed, frequently underutilized servers). “Systems consolidation,” according to Katzenburg, “has become a key focal point for SI—and a matter of policy.”
SI’s Data Center Systems Environment
SI long ago realized distinct operational cost advantages by moving to a centralized, dense model of computing. To achieve these advantages, SI uses several different classes of dense servers, including System z mainframes, high-end and midrange IBM Power systems, Sun UltraSPARC-based servers, and occasionally scale-up x86 servers (such as IBM’s X4 System x servers). SI also deploys HP and Fujitsu blade servers to handle Windows serving.
Each type of dense server runs different types of workloads:
• The IBM System z is charged with running highly secure, transaction-intensive, COBOL-based workloads. (CICS transaction environments remain a more efficient way to process transactions in a tightly coupled, dense mainframe environment than using a myriad of distributed servers and databases.)
• IBM Power and Sun UltraSPARC servers run UNIX application workloads; most importantly, SI’s OS Plus Portal application environment. SI could standardize on one system’s platform in the UNIX space, but has chosen to split its UNIX business to create a “healthy competitive environment” between Sun and IBM and to create a means to leverage acquisition costs between the two vendors.
• Windows x86 servers run select branch custom applications, a full suite of client applications, and provide terminal services.
Why Mainframes? The Sysplex Anomaly
SI’s data centers mix distributed servers with blade servers and mainframes. SI will soon have installed 30 IBM System z (mainframe) footprints in 11 Sysplexes (large, clustered environments). This deployment of mainframes is anomalous; IT service providers usually base their compute offerings on 64-bit UNIX architectures, including IBM Power and Sun UltraSPARC platforms or x86-based architecture. And this anomaly deserves closer scrutiny.