Windows 2000 Datacenter Server represents a new approach to buying, installing, and maintaining a Windows server. Buying a preloaded system from an authorized OEM and turning to that OEM for both hardware and OS support is a mainframe paradigm. As with a mainframe, Datacenter's price is high—well beyond the range of even the most expensive commodity Win2K server—so opting to migrate to Datacenter is a major investment.
As I talked to administrators at companies that have made this investment, I found a variety of reasons for implementing Datacenter and a variety of Datacenter hardware solutions. The motivations and implementations of two companies in particular—Health e Connex (HeC) and Cisco Systems—might mirror your enterprise's concerns. Even though you might not yet be considering a Datacenter implementation, an overview of these two companies' decisions might give you some insight into when and how Datacenter could address your company's scalability problems.
Health e Connex
"Data where it's needed, when it's needed. That's my mantra," declares Ray Pedden, executive vice president of HeC. As an application service provider (ASP) for the medical community, HeC provides financial software for insurance companies, managed-care organizations, third-party administrators, and medical providers. These customers need immediate, around-the-clock access to information.
HeC's business is founded on the ageless theory that if you build a better mousetrap, the world will beat a path to your door. "In medicine, the business process should be a straight-through process, but it's not. There are all these expensive detours," explains Pedden. "Traditionally, one out of three medical claims is sent to the wrong place. Half of the claims that are denied are never followed up to see what the problem is and whether the denial is valid. The average accounts receivable for a medical provider is 150 days, and most health organizations are spending $30 to $50 in administrative costs to collect a $20 copay. Our business is designed to eliminate those detours." However, the mousetrap to catch these troublesome mice requires powerful, sophisticated computer technology.
Each HeC customer accesses software designed specifically for that company's needs. Of course, different types of customers have different software requirements, so HeC must offer a variety of applications.
Insurance companies, managed-healthcare companies, and third-party administrators need to update and query their subscribers' information in the HeC database. In addition to checking subscriber information, these HeC customers need to process claims they receive from medical providers.
Medical providers also require a wide range of services from HeC. These providers query HeC about patients' eligibility and insurance information.The providers also ask HeC for information about whether patients are covered for procedures. HeC queries insurance companies and returns this information to medical providers. HeC also receives claims from medical providers and delivers these claims to the insurance companies. To handle patient payments that involve credit cards, HeC uses standard e-business credit card authorization processes.
HeC began with 10 servers running on a Windows NT 4.0 network. Six servers housed a Microsoft SQL Server database, and four servers provided software to customers. However, as HeC's customer base grew, so did the number and types of services that HeC wanted to offer. Pedden explains, "As we began to put together all the processes, I realized I could buy a whole bunch of servers and fill a large room with them, or buy an expandable box that would give me the I/O processing capability I needed." Pedden began looking for a system that could deliver the right software interface to each customer, provide excellent response time, and run a database that would hold all the necessary customer information.
To deliver customized applications to customers, the company settled on Win2K Server Terminal Services (in application server mode). To store customer data, HeC chose SQL Server 7.0. As a platform for the applications and database, HeC selected a Unisys [email protected] Enterprise Server ES7000 computer running Datacenter.
The ES7000 has 16 processors, 12GB of RAM, and a disk subsystem consisting of one ESM6700 disk unit and one ESM7700 disk unit. The system has four partitions, with four processors assigned to each partition. (For more details about the ES7000, see Forefront, "Datacenter: Up Close and Personal," page 31.) Unisys loaded Datacenter on the ES7000 computer and delivered the box to HeC's support company, GlobalCenter (recently acquired by Exodus Communications). GlobalCenter's IT personnel installed the software applications and tested them in a lab environment for 3 days. The Datacenter machine then went online, running parallel with HeC's existing servers. After 24 hours, when everyone was satisfied with the new box's performance, HeC stopped directing traffic to the original servers, and the Datacenter server remained online alone.
Two of the ES7000's partitions run Terminal Services in application server mode. These partitions boot from internal IPLs and hold the application software that HeC customers need; customers use Terminal Services Advanced Client (TSAC) to access the software. Datacenter's built-in load balancing maintains performance levels.
The other two partitions are database servers running in a clustered configuration, using Microsoft Cluster service, which is part of Datacenter. The clustered partitions boot from the ESM6700 disk unit and share 360GB of storage on the ESM7700 disk unit. The boot drives are RAID 1 (mirrored); the data storage uses RAID 1 for logs and RAID 5 for data, with a hot spare.
HeC also has two 2-way Unisys [email protected] Enterprise Server ES2024 computers, each with 1GB of RAM, running Win2K Server. These machines are the domain controllers (DCs) that authenticate users; the systems also act as DNS servers. The ES2024s act as a redundant pair to provide load balancing and failover. For administrative tasks, the ES2024s and the ES7000 run Terminal Services in remote administration mode. "This is all off-the-shelf technology. There are no rocket scientists needed here," declares Pedden, adding, "It's almost like going back to the stone age of computing; in effect, our customers are sharing time on a mainframe."
Pedden explains that the decision to house the Datacenter server at GlobalCenter provided bonuses that a location in HeC's computer center couldn't. GlobalCenter has a fiber network, and Pedden says that his customers are getting nanosecond reaction time. "When you get the level of power that Datacenter provides," he comments, "the only bottleneck is the pipes."
GlobalCenter also takes care of maintenance, backups, and other administrative tasks. GlobalCenter uses a generator-based power backup, so even power outages of long durations won't shut down the system. (GlobalCenter also maintains HeC's original servers and keeps them up-to-date; HeC can hot-switch back to those servers if necessary. However, Pedden says all evidence indicates that such a switch won't be needed, and HeC expects to drop this redundancy plan eventually.) The IT staff uses Terminal Services in administration mode to upgrade customer software and add new software as needed. Unisys technical support staff also uses Terminal Services in administration mode to work with the server. HeC's IT staff never sees or touches the box. "For us," jokes Pedden, "this Datacenter computer is an example of virtual reality."
HeC plans to add more applications to its customer services menu and is also moving to increase its customer base. The company has already introduced, or is about to introduce, expanded medical-provider services such as patient scheduling, billing, and collections; clinical and drug-interaction information management; prescription tracking; and other practice-oriented functions usually found in full-featured medical-practice software. To further protect sensitive medical information, HeC will soon offer customers the ability to log on using fingerprint identification. For HeC's insurance-side customers, document-imaging software will maintain and retrieve original employee enrollment forms, simplifying coverage confirmation.
Pedden says that down the line he'll probably buy another Datacenter server. "I would want another server, running far away, on the other side of the country, with a peer-to-peer relationship," he explains.
"But we can run a long time on one machine," he hastens to add. Pedden is working with Unisys to optimize the current system; HeC wants to tweak the operating environment for I/O and tweak the applications to take advantage of that increased power. "Unisys says they can get 200 to 400 percent improvement, even staying with 32-bit processors," he says. "I'll be able to handle 125,000 concurrent processes."
Mark Beaupre, systems administrator at Cisco, says there's a simple way to explain his IT department's mandate. "Everything we do is centered on availability and uptime."
Cisco maintains Windows, UNIX, and Sun Microsystems' Solaris servers throughout the company's international sites. Client logons are 60 percent NT 4.0 and 40 percent UNIX. A variety of software applications runs on all the platforms. "We want to provide an agnostic viewpoint for server platforms, so a business unit can come to us and say, 'We want to run this application,' and then IT will pick the best platform for the application," explains Beaupre.
Historically, Cisco's high-demand applications have run on platforms other than Windows. However, Cisco's manufacturing division recently selected a Windows-based application, i2 Technologies' i2 Demand Planner, which ties in with the company's manufacturing processes. Beaupre describes the application as "a very low-level statistical forecasting and modeling application that does a lot of number crunching." The application is both processor and I/O intensive.
Cisco's IT department installed the application on a 4-way Compaq ProLiant 5500 with 4GB of RAM, running Win2K Advanced Server. A second, identically configured server provided a lab environment in which to test changes in the software before moving the new procedures into the production environment. When system monitors reported that the computer was operating at 80 to 90 percent for CPU and I/O activity, Cisco decided that upgrading to a Datacenter computer for the i2 Demand Planner application made sense.
Cisco's IT department turned to Compaq for a Datacenter computer. Compaq, i2, and Microsoft worked together to match all the application's components and processes to Datacenter's standards. When Microsoft certified i2 Demand Planner for Datacenter, Compaq moved forward to fulfill Cisco's Datacenter order.
"We chose the ProLiant 8500, even though it has an 8-way limitation, because these computers are fantastic servers," reports Beaupre. To continue the company model of keeping one machine in production and another available for testing, Cisco purchased two identically configured ProLiant 8500s, each equipped with eight 700MHz Xeon processors with 2MB of cache and 16GB of RAM. Each server's internal disk is a RAID 0+1 volume with four 18GB disks, and an external disk on a Storage Area Network (SAN) is a RAID 5 volume with twelve 18GB disks and a hot spare.
The external disk space is divided into four active partitions: The C drive contains 8GB of disk space and holds the OS; the E, F, and G drives contain 35GB, 50GB, and 100GB of disk space, respectively, and each holds data. Some unconfigured drive space is still available on the internal disk. Processors run across all partitions.
After moving i2 Demand Planner into production on Datacenter, the IT department began monitoring the production system's performance. Using SNMP, the staff measures processor, memory, and disk utilization. The reports show utilization at about 10 percent for all counters. Beaupre calls the combination of the ProLiant 8500 and Datacenter "fantastic." The utilization figures reassure him that scalability won't be a problem.
Beaupre also praises Datacenter's service model. "The support paradigm is very appealing. We don't have to figure out whether we have a hardware or OS problem and bounce between vendors. The one-stop support at Compaq means I don't have to think about it."
Traditionally, Cisco doesn't install multiple applications on any server. Says Beaupre, "Single server, single application has always made sense to me. To host multiple applications on a single server, you need to make sure that application vendors understand more than how to avoid DLL Hell; they have to configure software for sharing a computer without causing problems. The operating system technology and the hardware technology have moved the cutting edge to a place where you can run multiple applications on a computer, but most applications are behind that cutting edge."
In accordance with this philosophy, Cisco plans to run only the i2 application on the Datacenter computer for a while. "However," says Beaupre, "the strength of the clustering in Datacenter means we'll probably expand its use and run more applications on that server. I'd like to get into the database area on the current Datacenter computer, for either SQL Server or Oracle. In the future, we'll probably add Datacenter Server computers for messaging and Web hosting."
When another application does run on the Datacenter server, Cisco's IT department will begin using Datacenter's features to reassign processors to processes. The staff will use scripts that monitor processor-to-process activity and automatically reassign processors.
But Beaupre says that to host multiple applications on one Datacenter server, he needs Microsoft to provide additional logons for Terminal Services in administration mode. All Win2K Server versions come with Terminal Services built in, and installing the component in administration mode permits two administrative logons.
At Cisco, multiple applications require multiple administrators, and more than two administrators will need to work on a server simultaneously. "My operational mentality is that whatever you use must be able to scale up and scale out," he says. "Datacenter Server provides all of that, but there's a bottleneck in the manageability."
Scalability on Your Terms
For both HeC and Cisco, scalability has been the primary motivation to move to Datacenter. For HeC, scalability means the ability to offer more software features to more customers. For Cisco, scalability means that the server system doesn't become a bottleneck for intensive use of a complex application. Datacenter clearly provides a solution for both of these scalability challenges.