Virtualization in Emerging Africa
If you listen to the marketing people from companies such as VmWare, Cisco and EMC (as I did just recently) they will consume tremendous amounts of energy and exhibit an abundance of ethusiasm to provide you with an endless list of why virutalization – and more specifically, why their specific virtualization products and technologies, are right for emerging African countries such as Nigeria. I will try to break down some of the key arguments.
Power It is proclaimed that Virtualization would reduce power costs, and that this would be specifically beneficial to Nigeria where most power is locally diesel generated. The reduction of power consumption will mean saving diesel costs, and in turn saving money. Not to mention the ‘Green’ factor…
Lets give this a little thought. It is true that most companies utilise diesel generators for most, if not all, of their power requirements but this got me thinking a little bit as to how you turn the good old refined fossil fuel into the very electricity needed for me to write this post. As I understand it, and I could be wrong, a diesel generator typically used in Nigeria is primarily for supplying a continuous power supply as opposed to standby or complimenting another supply to cater for peak times. This means (according to wikipedia) that the typical application is:
“a generator running a continuous unvarying load, or paralleled with the mains and continuously feeding power at the maximum permissible level 8760 hours per year.”
Ok what does the mean for the virtualization argument? Lets say we use virtualization to consolidate hardware or, for ‘power management’. By reducing the amount of servers that are switched on and are running in the data centre, you reduce the amount of power required. That is a fact. But…. you do not reduce the amount of power produced. The generator still runs at the same rate – that is why it is described as ‘unvarying’ and is providing the same amount of power that is fed into the same massive batteries and then inverted. Therefore, you are not reducing your power costs. You will still use the same amount of diesel. It seems that the only real benefit would be that by consolidating your hardware and reducing the amount of servers that are actually switched on, you would only be reducing the heat generated in the data centre – thus reducing your cooling requirement.
So now what do you do? Tell your boss that you are replacing the existing, perfectly functioning, generator that you have already paid an arm and a leg for, with two smaller generators to run in series so that you can switch on off during the night? I would very much suspect that the response would be along the lines of ‘go boil your head’…
What the pretty slide shows also fail to point out is that a server will consume more power when it works harder than when it is idle. So by my estimation, you are only really going to overcome the inefficiencies.
There can be a good case to have a virtualization strategy to save on power, but only if you are planning to purchase generators or, to build a data centre, then by all means it can make good sense to adopt a the strategy for power management. You could then decide, using virtualization architecture, to turn of X amount of servers during dark periods, thus reducing the power consumption. If these reductions amount to being able to turn off one or two generators, then it is a strategy well worth adopting providing your cost to implement will be less than your fuel savings over time.
Managment Another claim from said vendors is that with fewer physical machines to manage, your IT infrastructure, management becomes simpler. To a certain degree, I can almost agree with that statement. I say that because the definition of simpler is not clear. If you are managing less machines the exact same way you used to, then yes – someone’s job has just been made simpler. Instead of 200 physical servers needing power, cabling, cabinet space, installation, updates, patches etc – you may only need 50. But managing a smaller amount of servers completely differently can be a challenging task.
Firstly, if your IT infrastructure group is to offer “IT as a service” and you introduce ‘the cloud’ as your infrastructure strategy, you need to consider things like provisioning and the management of virtual servers. Just because they are not physically there does not mean you do not need to manage them. VMWare may provide a useable, fancy interface for data centre management functions, but they come at a significant cost and require staff to be trained and familiar with the technology to be able to use it effectively.
This introduces another set of challenges. Training and retaining staff in an emerging economy can be a very difficult business. If there are training courses available you will be very lucky to find it anywhere near your own country, often having to send staff to the USA or Europe. Then once your staff have now gained a whole new skill set, you need to retain them! This can often be your biggest challenge because your staff have now become very valuable and open to being poached by other companies.
You have now spent a considerable amount of money of software and training and you may not have a finished project if you are struggling to train or retain your staff. The alternative is to opt for an expensive third party – an “integrator” or “solutions partner”, who will then hang you by the short and curlies on long term support contracts!
Money The last major and key argument is that virtualization will save you money. This is a very bold statement and one which the OEMs do their utmost to try and convince you of. Vmotion, Vsphere and all these other tools and applications are not free. They are, in fact, very expensive. Cisco Nexus switches are also not free inside christmas crackers and when you bundle EMC storage into the picture, you will feel like you have been sandbagged. My opinion, based on costs to utilise virtualization with the above OEMs, was that it was not cost effective. Almost to the point where it was cheaper to continue purchasing and managing individual servers instead of consolidating and virtualising.
However, open source technologies, when utilised correctly can back up the argument that it is indeed cheaper to use virtualization than opting for traditional methodologies. I will give one example:
Let’s imagine that you have many clusters in your environment, specifically where the cluster is in an active/passive configuration. If they are built in a traditional 1 & 1 fashion, then you can say that you configuration is 50% efficient – 1 machine is active, 1 machine is idling on standby. If you have multiple clusters then you start to have multiple machines lying around waiting for something to happen. Using virtualization, it is possible to remove the hardware element from the architecture thus increasing efficiency. An example could be that you have four clusters but only five physical servers. Your infrastructure would be that eight virtual machines are created to make up the four clusters. The clusters will reside at any one time, on four of the five physical servers. This means that now there is only one machine that is idling as a standby, making your resources 80% efficient.
Utilizing technologies such as Xen, Redhat, dm-multipath, LVM and GFS (to name a few) then it is possible to build and configure a reliable, scalable virtualization strategy that is indeed fact cost effective. But my advice is that – with everything you do, you should always plan your strategy by knowing what your requirement is BEFORE determining which technologies to use. Too often I see organisations embracing a technology or a strategy before they even have a requirement.
What this invariably leads to is the need to create a problem to match your shiny new solution – which is exactly what the sales people want from you ….