February, 2008
by Todd Bourne, Theo Adis, CPT Global Ltd
Thanks to Large Scale Virtualisation technology available from software and hardware vendors today, it has become easier and easier to prove significant cost savings and to implement rapid application deployment processes in the IT organisation.
Unfortunately in many organisations Business Units are still averse to co-locating applications on the same physical server. With the current buoyant economic environment and the perpetuating opinion by some of our IT counterparts that "iron is cheap" it is still a challenge to convince the Business (who controls the purse strings) to go down the path of consolidation and virtualisation. This results in the continual poor utilisation of IT resources and Datacentre facilities and the spiralling costs of implementing and managing the exponential growth of servers in the Datacentre. This is much to the frustration for the business minded capacity planner and CIO, who is under continual pressure to control and even reduce the IT Budget.
This paper shows how to keep the Business Units and the CIO happy. It shines a Capacity Management light on the myriad of virtualisation and consolidation technologies now commercially and freely available with proven results in large scale server environments. Using a business orientated capacity planning approach, the authors show how the Business and IT can communicate effectively in order to achieve the same outcome.
The approach extends to Datacentre capacity planning (facilities, space and power) which has become a driving force to implement server consolidation and virtualisation, providing an End-to-End capacity planning approach to an organisation.
Background
In the past 12 months we have been involved in numerous Capacity Planning and Performance Tuning exercises involving large scale midrange server environments which have embarked upon the Server Consolidation and Virtualisation path. It has been very interesting working in these environments, finding out what different technologies and challenges have been faced. In many cases the challenges have been similar and in many ways they were very different. We asked ourselves what would I do if I could influence the IT procurement decisions in these organisations and what Virtualisation approach would I implement. We have incorporated these experiences, ideas and our colleagues' experiences into this paper.
Introduction
CPT Global prefers to use the term "Server Rationalisation" rather than terms such as Server Consolidation and Server Virtualisation. The terms Consolidation and Virtualisation conjure up many deep seated opinions and doubts regarding the question of how to drive up CPU Utilisation on midrange servers. Rationalisation shines a new light on this topic and it "hints" towards a more pragmatic approach.
Server Rationalisation conjures up a method that "makes sense", pragmatic and easier to implement as compared to CONSOLIDATION and which VIRTUALISATION technology is best. We don't really care what technology we use, we just want to make more effective use of our hardware resources i.e. drive up utilisation from the industry standard of 10% - 20%. World economies are buoyant and in some sectors are in "boom times" e.g. finance markets, resources and manufacturing.
With businesses doing so well, investment in IT infrastructure is accelerating to the point of putting huge stress on IT organisations to deliver applications and infrastructure. This not only applies to Production environments, but also to DRP, TEST, DEV and SVT. Combining the impact of "pent up demand" to overhaul aging infrastructure and the constant demand for more infrastructure, we are finding a huge pressure to roll out new infrastructure at a faster and faster pace.
In some cases it is much easier to replace an old server with "like-for-like" without considering factors such as "do we really need this app" and "can it reside on other servers". Due to the "glory years" of IT in the 80's and 90's when IT organisations had huge control over an ever increasing IT budget, the 00's have seen the reins pulled back and Business controlling the IT budget. IT is now seen as a cost centre and managed as such. Cost centres need to keep controlling and reducing their costs! As such, when talking to the Business it needs to be in "business-lingo" rather than in
"techno-geek" speak.
With all this IT infrastructure being rolled out we are now reaching the almost absurd situation of Datacentres running out of space and power. After the so called "demise of the mainframe", or more appropriately "optimisation of the mainframe" in the 90's, there was a world wide glut of Datacentre floor space. Huge mainframe, storage and SNA networking footprints were reduced dramatically. Nowadays the same Datacentres are filled to the brim with shiny new midrange servers, all busily humming along but with very little workload!
Meanwhile, the pressure is on to deploy new applications and associated infrastructure "just in time" with - in most cases - very little warning and involvement by the Capacity Management team.
Now that we have thousands of servers with even more OS images to manage, traditional Capacity Management Data Collection, Storage and reporting is too hard and time consuming. We need a new way of doing capacity planning for the Large Organisations that have embraced (or will embrace) Server Rationaliszation!
Observations
Ideally, IT runs as a business with its own budget and offers a utility service, much like electricity and gas companies. In reality, many Corporate and Large Government agencies see the Business holding the purse strings, with IT forced to spend money on a Project by Project basis. This does NOT provide much flexibility for IT to leverage shared resources. Another outcome of this project view in the IT budget world is each business unit wants to have their own infrastructure. They paid for it and therefore they don't want to share. This means all these new n-Tier systems going in have their own set of Load Balancers, Web Servers, Database servers, Application Servers, Message Queues and associated Firewalls and networking infrastructure. In many high profile applications this also means copies in DRP, SVT and TEST/DEV, leading to an uncontrollable spiral of cost!
When IT broaches the subject of sharing infrastructure utilising industry proven virtualisation and consolidation technologies, the business tends to say "it works, don't fix it - I make enough money and I am happy to cover the IT costs - I don't want any disruption to my service!". This is very frustrating for the business focussed capacity planner and for the CIO that is continually under pressure to reduce and/or decrease the rate of acceleration for the IT Budget.
Interestingly IT itself is also averse to encompassing new Server Virtualisation technologies. System Administrators are under constant stress to manage BAU and the constant flood of requests for deploying new servers. They don't have the time and people to look at new complex technologies and to plan for the future, even though in the long run it will make their job easier!
Architects and capacity planners are looking at the future whereas the Server Support teams are locked into fixing the issues of today.
Whilst the business has coped with the huge IT spend over the past couple of years, Datacentres around the world are running out of floor space and power. Real estate within the M25 area in London that can be converted to Datacentres are constantly under bidding wars with demand for power exceeding current supply by > 400%. Since lead times for fitting out and building Datacentres is at least 6 - 9 months, IT Managers are demanding better Capacity Management. What a fantastic opportunity for the Capacity Manager!
We will look at the Virtualisation technology currently available later in the presentation, which will show the amazing and proven technologies available today to implement Server Rationalisation. Combining this with the opposite force of the Business the CIO and Capacity Manager are both FRUSTRATED.
This paper offers an opportunity to help reduce this level of FRUSTRATION.
Many organisations are still rolling out the same Virtualisation/Consolidation solutions they rolled out 2 years ago. Even though this has provided a vehicle for rapid infrastructure deployment and a level of Consolidation, utilisation is still low! This combination of features is very favourable for the outsourcer (i.e. more money for the IT outsourcer or service provider).
In most cases virtualisation is still constrained on specific hardware allocations rather than applying more recent "soft" parameters utilising pools of resources. In many cases it would be much cheaper to deploy applications, even if they share OS's, on simple commodity servers rather than giving them their own partitions on an Enterprise multi-partition server. Some Enterprises are rolling out ALL applications on multi-partitioned servers with high availability backplanes, where in fact the application can either be rolled out on single image commodity servers or co-habitated on other existing servers. SUN Containers can also be used to isolated applications within the same OS image!
Again, Datacentre lead times are causing IT to start thinking long term, which is great news for Capacity managers. In outsourced environments the customer is now being dictated by the outsourcer to provide Datacentre resources as required in a timely manner. In many occasion the customer is forced to do its own planning and put pressure on the outsourcer to provide these resources. Cutting down the power footprint of the servers being rolled out will be a great benefit.
Capacity Planning Lessons
Capacity Management for IT Enterprises that span multiple Datacentres, possibly in many countries, deploying more complex Virtualisation platforms require a new paradigm in thinking. The new era is more complex, has many more servers (resource entities) and has to encompass IT expenditure "end-to-end" to include Datacentre planning. It also requires the capacity planner to write the business case Justification for the new server rationalisation platforms. Rather then re-invent the wheel let's learn from our past! George Santayana the poet said "Those who cannot remember the past are condemned to repeat it".
We have learned the lessons to enable us to move forward quickly and effectively. Effective Capacity Management requires the buy-in from the business and IT executive management. This is exactly the same as if we want to implement Server Rationalisation!
Simplify, simplify and simply by knowing your audience and documenting your target.
Aim to incorporate "simplication" into the Server Rationalisation approach (to be detailed later in the paper) to come up with a new approach to:
- Lifting Server Utilisation'
- Reducing server floor space and power (power supply and air-conditioning) footprint.
We have seen many advances in recent times in the field of business capacity management (refer to last year's CMG paper "More Tiers Less Tears"). Once the hard work has been undertaken to relate business workloads to Resource consumption, it becomes very easy to predict future resource requirements based on business drivers. When choosing WHAT to apply Server Rationalisation to, it is best to KEEP IT SIMPLE and CHOOSE YOUR BATTLES. The authors frequently look for the "path with the least resistance" as a possible starting point.
Many years ago IT projects were approved on a payback period of 3 - 5 years. In today's environment a payback is required within 12 months and MUST have recurring benefits into the future. Since IT is considered to be a Cost Centre, it pays for the capacity planner to be Cost Conscious. Being aware of all the cost in a basic TCO will go a very long way, hence the requirement to have a good relationship with the CIO and CFO!
KISS is very useful in many aspects of the modern business capacity planner. As the number of elements you want to cover increases, complexity also increase but in an exponential manner.
The Approach
Handling the opposing positions of the business and CIO and utilising the current Virtualisation technologies, we can provide a "virtual" dedicated hardware platform for each application and/or business service. Being seasoned Capacity and Performance Management professionals we can all estimate and manage the different workloads to ensure that we drive up CPU (and other hardware resources) up to an accepted Rule of Thumb (ROT) of 60% - and beyond! By utilising small partitions with shared pools of resources and workload management we can give business what they want and still leverage (expensive) shared resources.
Where there is a level of (hardware) virtualisation and CPU utilisation is still low, existing servers can be "squeezed" to provide additional capacity for new applications, thereby reducing the need for additional servers. This is especially useful for Datacentres that are already constrained. In our experience just by tuning the database can provide a 50% relief to server footprint!
In many situations it is NOT worth consolidating into a shared platform.
Focus on areas/servers that would benefit the most from consolidation/virtualisation. In some cases a stand alone server with a small application that is "chugging" away may not be a good candidate.
Decide on your Rationalisation Criteria!
Money, Money, Money. Since the business owns the IT budget and the CIO is under pressure to reduce it, use FINANCIAL terms and NOT TECHNICAL terms. Who cares about bits and bytes, the business wants to talk Dollars and Cents. "How much do I need to spend, what will I get for my money and how much will I save?"
No doubt that we have to get better and better in rolling out infrastructure for the avalanche of new applications, but we also have to get smarter and smarter. Datacentre Capacity Management (facilities, space and power) has become a driving force to implement server consolidation and virtualisation, providing an End-2-End Capacity Management approach to an organisation. Since we need 6 - 9 months notice to provide Datacentre facilities, we should ask for 9 - 12 months notice (AND involvement) for new applications.
Another aspect of the 00's is the amount of money available for IT infrastructure projects is that Businesses are undergoing massive Billion $ Transformation projects. Telstra in Australia is spending in excess of US$3B in its IT and Networking Transformation exercise. This provides a great opportunity to create a suitable Server Virtualisation and Rationalisation platform. With this goes the activity of Capacity Management, hence a perfect opportunity to design and build appropriate Capacity Management processes and solutions. Undoubtedly the Virtualisation and Capacity Management principles we come up with will influence the Design and Capacity Utilisation of future systems.
Where much of IT's BAU costs are related to aging underpowered servers due to the relatively high cost of software and hardware maintenance, we can avoid this in the future by imposing a 3 (to 4 year) Technology refresh for new projects/applications. We propose that nothing goes into Production unless the business unit running the project takes responsibility for the cost of performing a technology (hardware, software and even application) refresh.
There are (3) main categories of Server Rationalisation, based on the author's experience. This of course is not an exhaustive list but can be used a good starting point for the business focussed Capacity Planner and Manager.
EOL (End of Lease) are servers that are about to come to the end of their leasing period, and the company has to decide whether to extend the current lease and accompanying hardware/software contracts OR replace it with a new server and associated hardware/software infrastructure. The course of least resistance is to replace an old standalone server with a new standalone (single OS image) server. Even though the cost of the application on a new server will be CHEAPER than extending the lease/hardware/software licenses, it is NOT the best solution.
CPU consumption will be even LOWER (on the faster hardware) and the Datacentre floor space and power issues will be exacerbated. It will be MUCH better for the company to move the application or workload to a shared "virtualized" platform. If there are enough of these EOL servers about to reach EOL or are already on expensive maintenance contracts, it will make good/easy business sense to create a Server Rationalisation platform for all their workloads.
Existing Virtualised Platforms are ones which were implemented 2 or more years ago. Again these are approaching EOL and with 12 months of planning these workloads can be moved to new Virtualized platforms. These servers are mostly candidates for SQUEEZING, that is, tune and implement virtualisation principles that will allow more OS partitions on the same physical server. By squeezing we can maximise our investment in Datacentre floor space and power consumption by putting new applications/workloads onto existing servers WITHOUT having to share OS's.
Traditional concerns around Server Consolidation are related to Application sociability concerns. By supplying each application/workload with their own OS image we avoid these "religious arguments". This does NOT mean we can't leverage technologies such as SUN Containers where multiple applications can have partitioned OS resources BUT share the same OS. (Ideally we do both but if we want to address a legitimate concern the Business has we can avoid the whole discussion of sharing OS's).
New Virtualisation platform. Great place to start by providing a Business Case for a "Trial" platform. This would have all the features and configuration of your ideal Virtualised platform (micro-partitions, Virtual IO, Pooled Resources, Workload Management) that can be used for new applications and for moving workloads from retired servers. Once the trial is proven to be successful the IT organisation no longer has any hurdles in order to implement the virtualisation solution company-wide.
Some applications and systems are so big, they should NOT go on a general all purpose shared platform. They will be better managed and will most likely have higher availability if they have a dedicated "virtualised" platform. Still use the ideal features as above, but dedicate all physical servers to the one application or service.
This is the general approach for all Server Rationalisation exercises. It all starts with a review of the assets that are being considered for Rationalisation and then a high level Total Cost of Ownership (TCO) exercise. Identification of OLD servers and associated payments for maintenance (hardware, software, facilities management etc) may be a difficult exercise. In some cases physical audits were conducted and servers switched off to determine if they were still required by the organisations. This is where the KISS principle is very useful.
Make ASSUMPTIONS and fill in the numbers if you don't know what the exact costs are. Asking the vendor for their costs is a useful exercise as well, and reconciling with what IS known within the organisation. This approach is perfectly fine as long as the assumptions are documented in the business case and socialised with the IT management. It is no point creating a TCO with accurate financial data when the data doesn't impact the real cost to the organisation.
Companies may be Opex (Operational Expenditure) rich but Capex (Capital Expenditure) poor or visa versa. If aspects of the IT organisation are outsourced, the costing formula from the outsourcer is most important. Understand what the costing sensitivities are within your organisation and build a TCO accordingly. Again, KEEP IT SIMPLE.
Business Terms
Now that we know how to speak in business terms, the next step is to speak to the Chief Finance Officer (CFO), Chief Executive Officer (CIO) or Company accountant to explain how we want to save the company millions of dollars and reduce his future spend by millions per year.
Once we have the CFO/CIO/Accountant on-board, we write the business case and off we go!
This all sounds simple but as we have learnt in the past, "things are not always as simple as they seem". With a Solid Design Approach socialized with the Chief Architect and Designers, a simple but effective business case, a little encouragement by the capacity planner, server rationalisation can become a reality in your organisations. Setting the precedence to for ALL future projects will have a great impact and will make to job of Capacity Management easier. You have an opportunity to design and implement your preferred Server Rationalisation Design and associated Capacity Management Solution (will cover this topic a little later).
Virtualisation Principles
Virtualisation principles are similar across the different OS platforms, namely Windows, IBM-AIX, SUN-Solaris, HP-UX, Linux and zOS. We will cover the OS specific features in more detail in the next section; however these principles are explained in some detail below:
Workload Portability. Ability to dynamically move applications (processes) from one OS partition to another. This requires the files used by the application and associated software to reside on the same SAN storage, shared across all the OS partitions on the same server and other servers in the cluster (if clustered). It is an advanced feature and is used to dynamically move applications between partitions in the same physical server and OTHER servers, should the server run out of physical resources.
Virtual IO (VIO). Physical servers have a finite number of IO ports such as: USB port, Network Connections and Fiberchannel ports. VIO provides virtualisation of these resources across all the OS partitions to the physical IO ports in the physical server. This is the feature than enables many partitions (micro-partitions) in the same server. Traditionally a set number of Physical IO (PIO) ports are shared across a set number of partitions, thereby limiting the number of partitions per physical server.
Workload Manager is an automated process that looks at what resources are being consumed by each OS partition, and then allocates physical resources as required. As an example you can assign priorities to specific partitions in a physical server, and as higher priority images get busier, they can allocated more resources from the other partitions. A particularly useful example is the switching of CPU resources from one partition to another when processing switches from the On-line Day to the overnight Batch.
Rationalize or Compress Existing Virtualized Servers
This is an interesting approach for IT organisations that have already trodden the Server Rationalisation path AND it is time to incorporate more recent capabilities. In many of the older Virtualisation implementations hardware separation was the first step to the brave new world of Virtualisation. It was successful in providing a platform for rapid server deployment and proved that Virtualisation was a stable platform BUT without some of the more "soft" virtualisation features CPU utilisation was still very low.
Average utilisation (over a 12 hour peak on-line period) for one large European financial organisation was 20%, of which almost half (10%) was monitors. Instead of tolling out truck loads of individual servers the company was rolling out trick loads of multi-partition servers. This did NOT alleviate the growing issue of Datacentre floor space and power shortage.
The time is right to augment or enhance these environments to provide a platform for NEW applications, thereby alleviate the need for new Datacentre resources. These servers maybe sufficiently old enough to be EOL, in which they can be considered for a Technology refresh as per the previous section. Justifying investment for these type of severs follows the same path as for EOL servers.
Implement a Standard Virtualisation Platform for new Applications.
As per the EOL business case, you can help design and implement the preferred Server Rationalisation platform. In this case what we want to do is SQUEEZE the existing workloads into a smaller area within the physical server, thereby providing additional capacity in the existing servers for new workloads. These new workloads can also be from old EOL non-partitioned servers. Again, by SQUEEZING the existing Virtualisation platform we can offer relief for Datacentre resources and slow down the rate of IT expenditure for new projects.
This approach can be much less costly and less of a political "hot potato" than implementing a brand new (or different) virtualisation platform. There will also be an appreciable delay in demand for Datacentre resources, providing the IT organisation a window of time to properly plan for what Datacentre resources are required in the immediate future.
Going down this path you have the potential to drive down the on-going cost of IT on a PERMANENT basis. By implementing "software" virtualisation the IT organisation has an effective AND dynamic way to change the IT infrastructure to suit the business demands. Using capacity planning and performance tuning principles workloads and server resources can be managed to fit the business requirements by driving utilisation much higher whilst at the same time delivering an appropriate level of service to the business.
Taking a fresh approach to Server Rationalisation within the organisation WILL have great future benefits for an IT organisation. Here the capacity planner can assist the Architecture, Design and SysAdmin teams design an appropriate Virtualisation platform that manages the new platform from a Capacity and Performance perspective rather than from an "ease of system administration" perspective.
Another great benefit of such an architecture is the tight coupling between partitions in the same physical server. N-Tier architectures will benefit greatly from having high speed communication between the Database server and the Application servers.
Virtualisation and Consolidation Technology
VMware ESX Server
VMware has become the industry standard for Virtualisation on the x86 (Intel platform). Even though the license is expensive when compared to hardware costs, it provides unsurpassed flexibility in creating and maintaining OS partitions for new application. In Production, average figures are around 16 OS images. In Test you could have up to 100 images in which case not all are active at the same time.
Aspects of this technology has been has been incorporated into the UNIX world in recent times, which in turn was modelled on the mainframe virtualisation technology.
Currently VMware is ahead in many ways as compared to the UNIX offering, but UNIX is catching up. In some organisations the default platform is Intel running VMware and Linux or Windows. UNIX is only run for legacy applications. It is not too long ago when mainframe was legacy and UNIX was the new commercial platform for the midrange.
Hardware Level abstraction treats the Virtual Machine (OS image) the same as a physical machine: i.e. OS has no idea it is running on a non-dedicated server.
Hypervisor is wedge software that provides the virtual partitioning features running directly on hardware, under higher-level virtualisation software. Currently this is in software but chip makers like Intel and AMD are incorporating them into hardware (firmware).
VMware supports multiple OS's, Windows and Linux the most popular. Most people don't realise that it also supports Solaris86.
ESX has advanced automated Workload Management capabilities as well, which makes it a powerful virtualisation platform.
VMware started as an academic product based on the mainframe's virtualisation technology pioneered by IBM several decades ago. Server hardware technology has really advanced to the point where you can implement such mainframe features on the Intel (and AMD) platform. These types of virtualisation features have been mimicked by vendors such as Microsoft, who are releasing their version called MS Virtual server. Before too long these type of virtualisation functions will be incorporated directly in firmware. There is even a shareware product (written in Linux) called Xen.
SUN - Solaris Container
This has been around for some time and is a really strong Virtualisation platform. It differs to VMware in that it provides hardware level partitioning within the (1) OS image. Rather than partitioning a server and having a number of OS images (one per application) you can run multiple applications with their own hardware allocated (hardware separation) within the 1 OS image. This feature gets around the business concern of application sociability. It also allows you to 'reboot' containers separately, ensuring that a change on (1) container does not impact another container. It is only for UNIX and is available from Solaris 10 and above. There is some physical separation available in later versions, namely v8 and 9, but not really for earlier versions such as V2.3.
As you will see in the next few paragraphs, SUN has addressed some of these short comings with other virtualisation offerings.
SUN - Solaris Partitioning
This is more akin to VMware where you can run multiple OS images on the (1) server. You can run different versions of Solaris in each partition but the biggest draw back is that each partition has to span at least (1) CPU board i.e. 4 CPUs. The server running this type of virtualisation are at the Enterprise level so tend to be very big and mission critical. Australia's largest Telecommunications provider is under going a huge IT transformation project and they are deploying e25K's (>90 CPUs) by the 10's.
This is a very new feature available for the new SUN T1 and T2 servers. These are ideally suited as Web and App servers where high transaction throughput is required. CPU's are multi-core and each core has multiple threads. Each micro partition can be as small as 1 thread, but realistically (1) core which has (4) threads is the minimum.
IBM - Micro Partitioning
This is an extremely flexible platform. I call this VMware for UNIX. It has all the features we would want in a Server Rationalisation platform, alongside our VMware platforms. When comparing each Virtualisation platform, keep in mind that choice really depends on what the predominant OS platform is in your organisation. As an example, you wouldn't consider IBM micro-partitioning if the majority of your servers were SUN (and visa versa). If you are an AIX shop you would be "silly" not to run the all the features of this virtualisation platform.
HP - VSE (Virtual Server Environment)
HP had HP-UX partitioning on their Superdome Enterprise servers much the same way SUN has their partitioning on their large servers, with the same draw back i.e. each partition had to have at least (1) CPU board. With the advent of HPi11, HP-VSE now behaves more like IBM's micro-partitioning and SUN's recent Solaris Logical Partitioning. It too provides is with "VMware" on UNIX.
Capacity Management Approach
There is an opportunity to design a "no frills" but appropriate capacity planning, performance tuning and reporting solution for large scale midrange virtualisation implementations. Central to this solution is the Capacity Database (CDB) and easy data collection and database population. Key is to have a hand full of metrics and have a reliable method for extracting the metrics (through firewalls and Management LANs) and populating the CDB. The solution must be "productionised"!
Detailed data is not necessary. Half hourly or hourly observations are sufficient. It is always possible to turn on more detailed data collection for Performance Tuning issues when they arise. Consider a Data-warehouse model for the IT AND Business (Forecasts and Activity Counters) data, just like the Business has a Data-warehouse.
Add Datacentre centric metrics to the CDB. This will assist in the "very" proactive planning of the Datacentre resources. Plans for midrange servers, as well as other resources such as cabinets, network and SAN infrastructure can be fed directly into the plans for existing or new Datacentre capacity.
Opportunity to implement new processes to get Business buy-in for all stages of the Application Development SDLC:
- Feasibility
- Design
- Test/SVT
- Implement
End-to-end Capacity Management Approach
Again, use the need to plan for Datacentre capacity to get business engagement early on. Use the Window of Opportunity, reduce the footprint, and hence increase CPU utilisation for the foreseeable future!
Tie onto Application SDLC.
Summary
This is the general process for justifying and implementing an effective Server Rationalisation platform BY "shining the Capacity Management torch"!
It works and remember "you don't need to boil the ocean". Go for a service that is attainable and provide the most benefit for the Business and IT organisation. Find "the pain" and provide relief. If the numbers stack up you have a great opportunity to make a difference in your organisation.
Remember, we are also helping the environment by reducing the power consumption of our IT business.