The Data Center Divulged: Grids and Blades of Glory - Part 2

Posted April 1st, 2008 by Joe Pendry

This is part two of a two part series with Blade Watch editor Martin MacLeod. Here, Martin discusses the financial sector of IT, energy efficiency and green data centers, and the impact on performance when using virtualization to develop an energy efficient data center.

Part one can be read here.

StackSafe: You’ve worked in the financial sector of IT for some time. Our research indicates financial companies are more mature with regard to testing of IT changes – any thoughts on why that might be?

Martin MacLeod: The financial sector tends to be more risk averse, to want to deploy more tried and trusted solutions, though it does depend on the organization and the objectives. To implement a new technology needs approval from the various stakeholders usually within IT, the architecture and strategy teams as well as those that will support it. At the same time there tend to be approved vendors, approved tools for specific functions - the concept of a supported technology or a supported product, the objective being reducing the support cost, ensuring compliance, availability and reliability.

New technologies such as grid or virtualization tend to start off as development projects within IT, or on a business line basis and then evolved by the strategy and architecture teams to a bank wide standard solution. In the grid world for example what tended to happen was application team A, might write an HPC solution to meet a processing/business requirement, that code would be evolved until something else could be brought in that might achieve the same thing. The idea being that buying a product is good in terms of compliance, improves supportability of the platform and provides escalation paths in the event of an issue arising. By buying in my grid solution I ensure that applications are coded to a set standard, that they are industry standard, when a developer leaves I can continue evolving that code etc.

The other thing to consider is that bringing in new technologies might ‘change the way we do things’ which not everyone is ready for. If we take grid for example - who supports Grid? The Infrastructure teams? They will cry (possibly and rightly so), but that’s an application! One application team is unlikely to want to look after the actual Grid part, help application teams with their issues, as well as look after their own code development/support issues. With that in mind - do I need a new business unit, a new team to look after it?

StackSafe: You have blogged numerous times about “The Green Grid” and energy efficiency. How much traction and interest are you seeing in this trend?

Martin MacLeod: Energy efficiency seems to be one of these things that tends to split people within the IT industry. It tends to be related to the role that the person plays within the IT team. The CIO, the data center teams tend to be interested in energy efficiency, they issue and read the reports on where we are with the data center, how much power or cooling we have left - how long it is until we reach capacity or adopt a watt in/watt out scenario.

At the same time there are the teams where performance is their objective, where data center power is a data center issue, an IT thing, not what the business worry about. However, as the data center comes on the radar in terms of business risk, as the data center becomes a point of focus within your business strategy that will change.

We need to continue the momentum of the green message within the business context. Place a focus on deploying business transforming, or business enabling solutions in an energy efficient way. The concept that virtualization can lead to a more dynamic infrastructure, allocating resources to meet the business need, in doing so it can be deployed in an energy efficient way, reducing your energy requirements and the carbon footprint of the data center.

The green movement, the move to energy efficiency is going to come from three main angles. Firstly, consumer/external pressure to be more responsible in terms of your corporate social responsibility - we’ve seen this with the carbon rating/green rating awarded in the banking sectors. Secondly from the existing data center constraints, we need to continue to meet the business needs within the constraints of the data center in terms of power and cooling - how do we do this. Thirdly due to the cost of energy increasing, organizations are going to start (if they aren’t already) looking at how their data centers, how they can be more efficient - the cost of power is now the major operating cost of the data center.

StackSafe: What should companies consider when rolling out energy efficient solutions? How will they impact performance, complexity, etc.?

Martin MacLeod: The main thing to think about in terms of energy efficiency is the affect on the infrastructure or the application as well as the rate of return. As with anything, thinking about what those changes might mean to the different stakeholders involved with your data center; the business users, the IT and the data center management/facilities teams.

For example, if we implement DC power, what operational issues does this present, and what is rate of return - how long does it take me to make my money back?

A good example of this was a colleague who was telling me how they had decided to increase the temperature of the data center to 25 degrees Celsius. When asked about system reliability or hardware failures, it was explained that the hardware cost might rise by £100,000 a year, but the energy savings were £500,000 a year, the expected business risk of doing so was seen to be marginal as a result of their infrastructure.

In terms of performance, this is always an interesting one, choosing the energy efficient solution doesn’t necessarily mean it’s slower. Looking at the application or business requirements is key, at the same time looking at the total processing capacity required seems to be the way forward. For example in the blade field, choosing the mid-range processor power servers can provide you with the highest performance per watt ratio, choosing the fastest processor might give you marginally higher performance for a lot more power used or cooling required. You might find that buying 250 blades with the mid-range processors (from a power standpoint) gives you the same performance as 100 super fast blades, however the thing to note is the mid-range blades have less power and cooling requirements. The 65w processor as opposed to the 95w one.

In terms of complexity, consider what those energy efficient solutions might mean on an operational basis, switching all the servers to SAN boot might save you considerable energy, but consider your SAN capacity, the network/SAN infrastructure required to do so. Innovation needs to be done within your business and operational constraints, it’s often easier to sell a concept such as lower voltage processors and switching those storage arrays to SAN rather than going for a high cost or perceived high risk solution.

StackSafe: Anything new on the horizon for Blade Watch?

Martin MacLeod: On the horizon for bladewatch, I’m planning to start podcasts shortly, I’ve been organizing the server work involved and I’ve got to publish my guide to virtualization, which will be a pdf for download, covering best practice, hints and tips etc. I’m also wanting to do some more videos, ‘how to guides’, how to replace a hard drive that kind of thing, as well as more interviews.