All walking in the same direction

Last week I, like many others, attended the Uptime Institute Symposium in New York. It was great to be a part of an event which brings together IT managers, datacenter owners, C-level executives and anyone with an interest in server and datacenter management and operation.

 

The recurring theme this year was definitely energy efficiency and cost reduction and a large number of the keynotes and sessions (including my own) focused on reduction of power consumption through better power management, more efficient equipment and making the most of what you have; all themes that are close to my heart.

Unfortunately amongst all of the positive discussions, there was a roadblock that became more apparent as the week went on. There is a very clear divide between IT and facilities where the server and datacenter managers often don’t pay the power bill and to a certain degree, don’t care who does and the facilities team often have no input into the server provisioning process save but to provide space in the rack. This divide often makes if difficult for businesses to implement cost saving initiatives into the datacenter. The IT teams don’t always see thee benefits and the facilities team often don’t get to decide.

Technology is advancing and I’m hopeful that as power and datacenter space become more valuable commodities, this divide will be forced to shrink and the IT and facilities teams will start to work together to make the most of the capacity they have.

Heather Clancy wrote an excellent article on just this topic expanding much further on the points that I have made. Check it out here: http://www.zdnet.com/blog/green/power-play-green-efficiency-means-more-to-cfos-than-green-it/11794

Another way to think about ‘Useful Work’

1E’s NightWatchman Server Edition is all about visualizing efficiency by calculating ‘Useful work’. But what is useful work? There are many ways to explain it but I was having a ‘moment’ this morning and I came up with (I think) a good one.

If you like Formula 1 racing, you’ll get this. Imagine the systems management applications, anti-virus, backups etc (the non-productive work) as the pit crew. Of course they are needed to change tires, refuel and make sure the car is working effectively. However, as long as the car is in the pit lane, it’s not making up ground on the track, overtaking or finishing laps. Now imagine the car is always in the pits. It’s not going to win many races that way and if its not winning races, the pit crew aren’t needed anymore.

In other words, if the only thing a database server does is spend its life in the pits, you might want to consider either firing the pit crew or better still, removing the server altogether.

Useful work….there it is…..

The Bad Kind of Sprawl

When I hear the word ‘sprawl’, I always think of stretching out by the pool on a lazy day in the sun, clutching a mojito or lying in bed late on a Saturday. That is the good kind of sprawl. However, there is an altogether more sinister use of the word. I am referring to the emerging issue of virtual machine sprawl.

In recent years, a number of new technologies have emerged in an effort to slow the growth in the number of servers and to make server provisioning, management and monitoring easier and more efficient. The most prevalent of these new technologies is Virtualization.

In simple terms, virtualization is the concept of containing many ‘servers’ within a single physical box. Imagine putting shoeboxes within a single, larger box. Anything you place in the shoebox will be inherently contained within the large box but will be separate from something located in another shoebox. This idea is effectively how virtualization works. Each ‘virtual server’ is allocated a portion of the physical machine and is separated from both the physical hardware, the operating system running on the hardware and other virtual machines by a layer called a ‘Hypervisor’. The basic concept of virtualization is shown in the image below:

Virtualization was designed with the intention of reducing cost and complexity of provisioning and maintaining servers in the datacenter. Firstly, you can have many servers running on a single piece of hardware, and since space is a valuable commodity in the datacenter and cooling is expensive, it reduces the demand on floor space and infrastructure. Secondly, since there is no need to provision hardware every time a server is needed, the process for building a new server is greatly simplified and accelerated. Also, management of these virtual machines is made easier using proprietary tools that are often provided with the virtualization solution such as VMWare’s vCenter and Microsoft’s System Center Virtual Machine Manager.

However, with any new technology comes challenges, and whilst virtualization has been hailed by some as the ‘silver bullet’ to resolve power and space concerns in the datacenter, it brings with it its own inherent problems; one of which is virtual machine sprawl.

Virtual machine sprawl refers to the result of poor provisioning and management of a virtual environment where many virtual machines are created and configured, and then are simply forgotten about. The environment sprawls out of control and the snowball effect takes over. The more virtual machines that exist, the more difficult it is to manage and find unused virtual machines. Instead of re-using machines, another is created and the cycle perpetuates. This is virtual machine sprawl.

Of course, this doesn’t happen overnight but the biggest concern is not that it is happening, but that many users of virtualization are simply not aware of it. So the next question is ‘How does it happen?’

Virtual machines are quick to provision and often at seemingly ‘no cost’ to the application owner and end user. This mindset leads to virtual machine sprawl. If we compare the process for provisioning a physical server with that of creating a new virtual machine, the situation becomes clearer. Provisioning a new physical server is often laborious and time consuming and can look similar to the following:

1. Justify purchase of server
2. Raise PO
3. Place hardware order
4. Hardware manufacturer builds/picks server
5. Server shipped to datacenter
6. Unpack server and rack
7. Patch network and power and run cables
8. Build Operating System
9. Patch Operating System and install management software
10. Hand over to application owner

As you can see, there are a lot of steps here and this is by no means an exhaustive list. In addition, some of these steps can take a considerable amount of time to complete, often resulting in the process taking several months. Not to mention the added complexity of finding space and power in datacenters that in some cases are nearing capacity.

Now let’s compare that to the process for building a virtual machine:

1. Create Virtual Machine
2. Take coffee break
3. Hand over to application owner

Of course, this list is being somewhat flippant but at the same time, this idea that virtual machines can be created quickly and at little or no cost often leads to virtual machines being created for testing, application development and expanding production applications and not being repurposed or deleted when no longer used. This, in turn, leads to an increase in management overhead, increased hardware demands on the physical hosts and unnecessary licensing costs for machines that are not in use.

If virtual machine sprawl it’s not addressed, it can have catastrophic impact on cost.Boeing recently commented to Gartner that they believed that their savings through virtualization could be entirely wiped out if virtual machine sprawl is not controlled.

1E’s NightWatchman Server Edition helps to visualize virtual machine sprawl and helps companies to gain a handle on the problem, enabling them to realize the greatest benefit and ROI from their virtualization solution through efficient use of technology.

-- This article can also be found on the 1E Tech Blog --

The $20 Million Server

Space is becoming more of a valuable commodity in data centers. With the lack of affordable ground space, over-stretched power grids and tightening efficiency legislation, building a new data center is often not an option. Consider the New York area where I live. Clearly there is a huge demand for high-tech and this means servers and data centers, however, whilst there is the infrastructure to build a data center on the banks of the Hudson, the area lacks both the physical space and the grid capacity to support one. In fact, the New York city metro area is significantly over capacity at present which has caused a noticeable impact on service, most severe on August 14th in 2003 when the already strained power grid was brought to it’s knees by a suspected system failure somewhere in Ontario, Canada. Generators from the North-East United States; including New York, tried in vain to shunt power westward to the affected areas causing a widespread blackout of Ontario, the North-East US, and parts of Michigan and Ohio.

To further confirm the observation that the New York metro grid is strained, it was reported that in the summer of 2004 New York City’s peak energy load was 11,150MW whilst the physical capacity on the grid was a mere 8,940MW and in addition, it is known, that due to constraints with the transmission lines, the vast majority of power for New York City must be generated within the city limits during peak times which does not provide a great deal of slack. Worst of all, the New York City and Long Island zones' electricity generating infrastructure has the highest average age of generating units in the state and is still highly dependent on an aging fleet of combustion and gas turbine plants bringing its long-term reliability into question.

These factors all combine to produce the $20 million server effect or the one server that simply cannot fit in a full-to-capacity facility and therefore is the catalyst for building a new data center. A large number of companies are very reactive in terms of their capacity and often, due to a combination of cost, complexity and sometimes I’m sure, an element of denial, new facilities are built on-demand rather than in advance. Instead of allocating budget, sourcing planning permission, carrying out the design work and breaking ground on a new data center before the extra capacity is needed, often projects that require new infrastructure tip the balance of capacity over the edge and the building of a new data center becomes a knee-jerk reaction instead of a proactive exercise. The $20 million server becomes a reality.

So how can this be avoided? Considering the challenges of actually locating and building a datacenter, making more efficient use of what you have is, in many cases, a much better and more cost effective option. Technologies such as ‘virtualization’ and ‘cloud computing’ can help to relieve the burden on over stressed facilities. In addition, gaining visibility into the consumption and efficiency of servers is an important component is building up a picture of wasted resources. Knowing whether or not a server is ‘useful’ enables the business to make more informed decisions about reallocation of hardware to reclaim precious space and capacity in the data center.

1E’s NightWatchman Server Edition visualizes server ‘useful work’ and is the seek and destroy tool for virtual machine sprawl...say goodbye to the $20 million server and hello to a more efficient data center.

-- This article can also be found on the 1E Business Blog --

Welcome to my new blog!!

Becoming ‘Green’ is something more and more people are starting to do. There is a lot one can do as an individual and companies can also become more green through efficient processes and systems.

‘Green IT’ is concerned with making an organization’ Information Technology as efficient and carbon neutral as possible whilst still ensuring it is able to effectively service the business.

This blog will focus on both technological and behavior changes in IT and aims to be a reliable source of information on Green IT initiatives and solutions.

Enjoy!

Andy