Home > Cover Story
 Print Friendly Page ||  Email this story

Shopping for Servers

Buying servers may seem like risky business, especially since it involves big spending. But by taking the right steps IS managers can make cautious decisions and choose the right solutions to ensure return on server investment. by Brian Pereira

IS managers are asking different questions about servers these days. And server vendors no longer pitch their products on the basis of tpmC, throughput, OLTP performance and hardware specifications. (These parameters are discussed later in the procurement process.) Today, instead of buying excess computing capacity to address future requirements, IS managers are more focused on immediate short-term requirements. Indeed, the agenda for server procurement has changed dramatically.

The need for servers begins with defining the business requirement and the applications to fulfill it. Then other factors, like number of nodes/users, availability, redundancy, scalability etc come into play.

Anil Valluri, Director, Systems Engineering, Sun Microsystems India says server sizing is a complex exercise and there are no fixed guidelines. "The decision depends on workload mix, transaction type and mix; user duty cycle, usage transients etc."

Valluri says the thumb rule for server sizing is to start with online applications, then look at the exact nature of the transactions (complexity, application logic etc), and arrive at the CPU and I/O utilizations for a required response time. He says one should look at the transaction mix and usage patterns, and arrive at aggregate CPU and I/O requirements.

The IS managers whom we spoke to, say immediate business requirement comes first. Instead of buying excess server capacity for the future, they'll buy just enough to address immediate requirements. And they'll buy a system that can scale up—one that can be upgraded as and when required.

Harsh Kumar, Advisor-IT, Hindustan Petroleum Corp. Ltd. (HPCL) sums it up well. "You look at the business need which is to be satisfied through an application. Then you determine what server that application needs. The server decision depends on what you are going to run on it, and what features are required specifically for that."

Kumar (and other decision makers at HPCL) take a different approach for buying server infrastructure. When talking to vendors they initially specify the expected server response time. They talk about the application(s) that will run on the server and the number of clients that will connect to it. Various load scenarios are presented, based on application usage. And nobody's talking about server components or specifications yet.

When additional server capability is required IS departments consider three options: reallocate resources, upgrade the existing server, or replace it with a new one.

Says Hilal Khan, Head-IT, Honda Siel Cars, India Ltd., "In case of upgrade, we consider the cost for the upgrade vs. planned use and the desired results. We also consider the cost of the upgrade vs. the cost of a new server. The level of the upgrade is also an important factor."

Server investment is huge; after all it isn't just a hardware box that one's buying. One has to factor in the cost of operating system and application licenses. The cost of a mainstream ERP package, for instance, can run into several crore of rupees. So the decision maker gets just one shot at it, and then the server must be used for at least five years. It must pay for itself during that period. Naturally, IS managers go out to buy servers with a long wish list tucked in their pockets. Once they decide on the applications, workload, and number of users, it's time to pull out the wish list.

S.R. Balasubramaniam, VP-Information Technology, HDFC Bank, puts scalability at the top of his list. "After considering the type of applications and number of users, I ask about the scalability. With a Sun E 12K I can start with four CPUs and then keep on adding processors as my requirements increase—going all the way up to 36 CPUs."

Like all other IS managers, Balasubramaniam is also concerned about ROI (Return on Investment) and TCO (Total Cost of Ownership). "The business that is running out of a particular server should be profitable—the server should pay for itself. The SLA (service level agreement) is important too," he says.

The wish list goes on to include other requirements like redundancy, availability, compatibility/integration with existing infrastructure; ease of management etc. Many of these requirements are driven by trends like server consolidation.

Vendors know what server shoppers want. In these times of constrained IT budgets, everyone's looking for the best deal: how many of the requirements can be fulfilled for the best price? Vendors talk about how they can help enterprises save costs. They no longer sell 'products.' It's only talk about 'solutions' and 'services' to 'help enterprises grow business.'

It may seem that the biggest customers for servers these days are those from the BFSI sector (Banking, Financial Services and Insurance). Companies like HPCL and HDFC Bank, which set up data centers within their own premises, also buy servers every six months or so. Server vendors regularly visit such companies and are frequently talking to IS managers. We spoke to some of these vendors and asked them about the issues that come up in these discussions.

Durgadutt Nedungadi, director, Business Development and Solutions, Industry Standard Servers, HP Asia Pacific, says managers are worried about the increasing costs of managing servers.

"Over the years enterprises have deployed several discreet servers for various applications within its infrastructure. Enterprises tell us that the cost associated with maintaining these servers, and the pain associated with it is too expensive. So they are looking for ways to simplify server management," says Nedungadi.

One way to do that is to change the administrator-to-server ratio. Typically, one administrator manages 4, 6 or 8 servers within a data center. Nedungadi informs that managers want to change the ratio to 1:20 or 1:25. Modern network infrastructure management software and techniques like server consolidation and server partitioning can make this possible. In fact, the Systems division of a major financial institution that this magazine recently visited, employs just four full-time network administrators to manage its nationwide WAN.

"Another concern is that enterprises want the ability to buy exactly what they need today and then scale their resources as their business and requirement grows," says Nedungadi. "They'll buy the exact capacity that they need today, and add more as and when their business grows."

Vendors also observe that enterprises are trying to make optimum use of what they already have. For instance if an enterprise has a payroll server that's used only towards the end of the month, and a more critical server that has high workloads throughout the month, they'd want to use the payroll server (when it's idle) for that critical application (for load balancing). So administrators look for tools that can do server redeployment and reprovisioning. And server vendors are providing such tools as value-adds.

Buying servers may seem to be a risky proposition as it involves big spending. But IS managers don't have to feel like punters placing huge bets at the racecourse. By taking the right steps (see box: While shopping for servers) one can make a cautious and suitable decision.

Meta Group analyst Phil Dawson—who is international program director for Server Infra-structure Strategies—says the first thing to do is an assessment of the infrastructure that's already in place. Then one should consider the business need to fit the application. Finally, one should measure the transitional elements of migrating to the new platform.

"The biggest thing managers are missing is the cost of change. They should factor in things like software licenses, required hardware etc," avers Dawson.

Dawson advises managers to consider the number of applications, and to stop worrying about scalability—since most platforms are scalable. "I think workload management and measurement is more important now."

And here's some more advice from the people who deal with servers:
M. Ganesh, Country Manager, Enterprise Systems Group, IBM India says, before making investment decisions, organizations must develop clear mutual metrics—estimates of costs, benefits, flexibility and risks—that unite technology solutions with business goals and strategies.

Pallab Talukdar, Director, Business Critical Servers, HP India, advises IS managers to think long-term and choose a vendor, platform and server accordingly. "IS managers need to ask questions about roadmaps, adaptive infrastructure, balanced performance and the number of applications that can run simultaneously."

Sun's Valluri cautions against server benchmarks. He says benchmarks like TPC-C for sizing are done in environments that are not related to customer requirement. "Customers should insist on performance testing of their own applications or do a comparison based on a set of real world benchmarks for SAP R/3, PeopleSoft Oracle etc. These real application workloads give a fair idea as to how the servers will stack up. Then the conclusions should be factored in with batch loads and the unique characteristics of customer environments."

In light of constrained IT spending, managers need to take a cautious approach when procuring servers. Before considering new servers one should think about reorganization or redeployment of existing infrastructure.

Techniques like load balancing and consolidation should be considered too. Various load scenarios should
be simulated in an enterprise environment, with the real-world applications.

Do not overlook other related factors like server room environment, server operating system and applications. The vendor is likely to prescribe a set of ideal conditions under which the server should operate. Some managers outsource areas like server room facilities and server monitoring. Also plan for future expenses on software upgrades, spare components etc—this is better known as TCO (total cost of ownership).

When making purchasing decisions one should always keep return on IT investment in mind.

What managers and administrators look for ...
  • Enterprises are more focussed on immediate, short-term requirements.
  • Business requirement and applications come first. Can the server fulfill this requirement?
  • Servers should be scalable or upgradable for future requirements.
  • Managers look for ways to reduce the cost of managing multiple servers.
  • Ease of management.
  • To make optimum use of resources, administrators use tools to redeploy or reprovision servers on the fly.
  • Long-term support (SLA). Does the platform have a long roadmap?
  • How does the server respond to various load scenarios?
  • How well does the server fit in with existing infrastructure?
  • Total cost of ownership (TCO) and Return on Investment (ROI).
  • Server availability and redundancy.
While shopping for servers …
  • Match your business application and the application workload to a specific server type/platform.
  • How can your business leverage on the server's features and specifications?
  • Ask the vendor about the server (or processor) roadmap.
  • How well will the new server fit into existing infrastructure?
  • Check response time, scalability, reliability, and availability factors.
  • Calculate TCO and cost of future hardware and software upgrades.
  • Factor in cost of the operating system and applications.
  • Check ease of deployment, redeployment, reprovisioning, and management/administration.
  • Ask the vendor about server room environmental conditions (support infrastructure).
  • If possible, test the server in your own business environment, with your own applications. Simulate various load scenarios.
  • Ask the vendor about support services, the SLA, and value-adds.

Brian Pereira can be reached at brianp@networkmagazineindia.com