Proven server technologies for large data centers, by David Geer

Contact me at david@geercom.com, or at 440-964-9832 (Fax:440-964-2172).

Data Center Management magazine

Proven server technologies for large data centers

Gizmos and gadgets for the data center

By David Geer

Server consolidation forces vendor competition, which produces better, more cost-effective technologies for the future. Determine how the use of server rationalization, virtualization, and popular operating systems can help save money and space.

By David Geer

Server environments are prime examples of the amount of change that can occur over a two-year span. By the turn of the century, data centers had over-purchased server capacity; today, they are streamlining.

When servers are added or replaced, mainframes are still the most frequent choice for core applications, while quick, cheap, modular servers are being tested on non-mission-critical applications.

Server Hardware

Data center managers are holding on to their legacy mainframes and seeking increased utilization.

However, according to John Rowell, vice president of operations/engineering at OpSource in Chantilly, VA, the mainframe's market share is declining, with high-end Unix servers like SUN's Sun Fire 15K and HP's 64-processor Superdome Unix server gaining daily ground.

Corporate data centers are examining TCO for Unix to rein in expenses, in some cases switching from other platforms, while low-end Sun servers - the E220 servers and the Netras, for example - are still carrying a strong presence in non-mission-critical applications.

Rationalization

The growth of certain vendors in the marketplace should not overshadow the fact that most data center managers are not buying new server technology. The focus is on deploying tools for infrastructure rationalization.

Rationalization applies monitoring and visualization tools to determine where servers are under-utilized, and to model techniques to increase utilization of idle processing cycles.

Rationalization helps to blueprint consolidation efforts, cut costs and power consumption, and improves efficiency. Through rationalization, different systems from multiple departments can be consolidated.

If data centers do invest in new servers it is often because the cost/savings benefits of buying into cheap, modular technologies – such as blade servers - are greater than those of reconfiguring current hardware.

According to Rowell, rationalization of server hardware and software saves up to 45% in ongoing maintenance and support costs, with an average figure of about 30% savings; data centers “can certainly expect to save [at least] 15% to 20% in 95% of the organizations,” explains Rowell.

Virtualization, Grid Computing

One tool for realizing consolidation of rationalized servers is virtualization. Virtualization runs applications on separate virtual servers within a single machine to increase utilization. This produces monetary savings with regard to time, power, and equipment.

Another useful consolidation tool is grid computing, in which jobs are shared across servers in a grid. Grid computing not only offers the ability to share jobs, but to prioritize them and shift compute cycles from one job to another – again, creating savings.

Each of these technologies can transform multiple racks of disparate, under-utilized servers into fewer racks of well-used machines. The resulting topologies often include combinations of new or legacy mainframes with 1 and 2U servers running Linux or a Windows Server OS.

Operating Systems

The most popular operating systems include Unix, Windows Server (2000, 2003), and Linux. Various iterations of MVS still abound on older mainframes.

Windows Server 2003 enables clustering among eight servers, increasing operation of compute cycles across clusters (grids). Linux has been adapted on IBM's new z990 architecture to support hundreds of virtual servers in a single mainframe.

The Linux/mainframe marriage enables multiple applications and work prioritization on one machine. Because Linux is a true, multitasking operating system, it can divide itself into several virtual servers, each running Linux.

“Not only is it a great consolidation platform [that] lowers operating costs, but it maintains the flexibility to handle spikes in demand,” says Pete McCaffrey, director of zSeries product marketing at IBM in Poughkeepsie, NY.

In the effort to trim an overabundance of Intel, Unix, and mainframe solutions, some large data centers are deploying Linux virtual servers on mainframes for critical and non-critical applications. This enables these centers to subtract from the remainder of their infrastructure.

However, a data center server trends survey in June 2003 by Richard Fichera - an analyst at Forrester Research in Cambridge, MA - found that 95% of data centers have no plans to abandon the multiple-server operating systems they have in use.

While approximately 98% of data centers use one or more Microsoft OS, 80% use Unix and approximately 45% use Linux. Unix, however, still holds the largest share of the market.

Server Types

Streaming video servers are notable additions to staples like databases and Web, E-mail, and FTP servers, as assessed by Christopher Faulkner, chief executive officer at CI Host in Bedford, TX. Demands for streaming servers have been on the rise for over a year. Real (Seattle, WA), QuickTime, Halo (Sydney, Australia), and Windows Media 9 are Faulkner's top sellers. Drivers for streaming servers include Web radio, video, online training, and video conferencing.

Common server makes and models appearing in the data center still include Exchange for e-mail, SQL databases, and Cold Fusion for Windows apps and Apache (Forest Hill, MD) for Web servers. Apache for Windows is now available and in marked demand.

Sample Environment

ANET has a new 10,000sq-ft data center in Chicago, which uses server clustering for distributed computing. Servers run Microsoft Windows Server operating systems from NT on up, and Linux. “We have used combinations of established IP networking, file sharing and clustering protocols such as NIS, NFS, yp, DNS/BIND, LDAP, Radius, ODBC and SQL. [We've done this] to create a server environment that supports applications across operating systems from a stable, secure location with redundant facilities onsite,” says Janet Rogers, chief operating officer at ANET, Chicago.

“ANET uses a battery backup system that keeps power levels stable during any temporary outage, and that system is further backed up by a diesel generator which can power the facility indefinitely,” says Ms. Rogers.

Significant amounts of ANET's data storage technologies consist in a series of network appliance storage devices, each offering up to one TB of RAID 5 storage capacity. Offsite storage backups are entrusted to DLT tape libraries.

Applications

Discovering the most proven server applications at any data center requires an assessment of its customers' most frequent application requests, which can vary widely from center to center. Most share a common ground in IP, with database applications leading the pack in meeting customer demands.

Monitoring and visualization tools rank high (for rationalization). Corporate and enterprise applications are increasing over Internet and hosting apps. Data center customers frequently develop their own applications to target their unique needs, requiring data centers to support them.

Customers also request suites like PeopleSoft (Pleasanton, CA) and SAP (Walldorf, Germany) where tighter integration with other business processes is desired. Linux has introduced software applications paralleling the breadth of those available under other platforms like Windows Server 2000, 2003.

Data centers are on the hunt for fast, Web enabled applications. “We're talking about databases, business intelligence applications, ERP, e-commerce facing apps - a portal or a traditional B2B, B2C e-commerce, or point and click,” says Steve Johnson, chief information officer at Optimus Solutions in Atlanta.

“We are seeing an increased demand for load-balancing, redundancy and advanced services,” says Ms. Rogers. ANET has also seen growth in voice, VoIP, and video applications.

E-mail is a critical application. Exchange 2003 is a popular, reliable package, particularly because it clusters.

Meeting Challenges

In the trenches, the latest iterations of Microsoft Windows Server and Microsoft Exchange are proving themselves reliable in clustering environments. An older Microsoft Exchange platform on an Intel single server failed at a leading hospital. Its 2000 users were without e-mail communications for three hours.

Optimus Solutions responded to what turned out to be hard drive controller failure with a complete package. Optimus informed the hospital they needed to move to Microsoft 2003 and Exchange 2003, which has clustering technology built into it. The upgrade included two robust Intel servers for clustering, and migration to active directory. The result was continual uptime through redundancy, even under heavy e-mail traffic.

Evolution

Code-named T-Rex, the new z990 mainframe is reaching out to Java, the Internet, and other technologies to compete with servers that already embrace them. The new mainframe houses up to 64-processors, each 40% faster than the z900's, making the 64-processor version six times faster than the z900.

Server technologies have improved at a marked rate over the last two years. “A lot of companies, even post September 11th, have driven a lot of dollars into R&D. It has shortened the development cycle a great deal,” says Mr. Faulkner.

Due to savings opportunities available through outsourcing to India and to the desire to renew faith following the dot-com bomb, vendors are bringing out servers quicker that are cheaper, smaller, faster, and more reliable.

Future Technologies

Larger mainframe customers have cut back on physical infrastructure, virtualized their servers, and gained a more responsive server environment. Similar improvements should proliferate across this space over the next several years.

In years to come, hardware and software vendors and developers will be focused on grid computing. The result will be heavy competition for current technologies from field replaceable servers that make up server grids.

Processing cycles will constitute an on demand utility where customers pay for what they use. Institutions like universities will be able to offer their idle cycles to communities and organizations during low use periods like evening hours, says Mr. Johnson.

Blade server adoption for Web and mission critical applications will force stiffer competition from Microsoft with increased Windows Server features and functionality. Unix will become more open source to compete with Linux.

Once consolidation is addressed and blade servers have established trust for critical applications, data centers will purchase the best new technologies for redundancy, reliability and resiliency. Individual drives will be replaced more and more with storage area networks (SANs).

“It's more cost effective to go with a SAN, to have all the front end web servers connecting to one giant storage area network. If a drive goes out, servers don't go down, it moves that data over to another partition,” says Mr. Faulkner. The individual server as a point of failure becomes less an issue.

As SAN and network-attached storage (NAS, separate storage with its own IP) become more cost effective, it becomes easier to justify clusters for increased reliability.

Ebb, Flow

By making do through rationalization, virtualization and other techniques, large data centers force server vendors to compete more intensely, producing much better products. When a new plateau is realized through consolidation, a resurgence of build up will ensue carrying with it the reward of resilient, cost effective data redundancy.

Sidebar(s)

Six Steps to Server Rationalization

1. Find a governing party within the corporation. A lead, control person that has the authority to determine how assets that may be spread across operating systems and multiple business units can be consolidated.

2. Audit the infrastructure. As simple as it sounds, many corporations do not know 'what they have'.

3. Determine the departments utilizing the infrastructure, what applications are being run, uses of the applications.

4. Determine overhead, operating costs and strategic areas to rationalize.

5. Create robust SLAs to the corporate stakeholders that the infrastructure will be supporting.

Determine future strategic investments...don't let consolidation shift the focus from ways to build future efficiencies through capital expenditures.

- John Rowell, VP, operations/engineering, OpSource

Check and Double Check

We always double-check our work. It's a pre-printed list of things that we always double check.

We make sure that we're, one, following whatever customer request is made upon us (it doesn't need to be this customer [speaking of Microsoft] it can be other customers). And, we always double check our work and then submit that to the customer, letting them know that this is the checklist that we follow and that they should let us know if they want it changed or modified.

So there's never any question about how the customer is approving this or how this is passing muster.

Kevin Howard, senior operations director responsible for work with Microsoft, OpSource