If your organisation finds it needs to process more and more data, you may have come to a crossroads. Buying in all the necessary processing capacity would embroil you in a new hardware maintenance cycle, and probably still wouldn’t provide the flexible scalability now needed – unless you buy a lot of extra capacity you’re almost never going to use. If you think cloud-based infrastructure services (IaaS) might have the answer, here are some considerations to help you find the right kind of provision.
1. The Provider’s Business Model
This might not be immediately obvious, so consider who else is using the service you’re looking at and for what sorts of activities, to see whether it’s likely to be a good fit for your business. If it’s a large, mainstream provider offering large scale at great prices, ask how consistent the numbers quoted are and what guarantees, if any, are in place.
Volume services and low prices often mean a sacrifice somewhere along the line. This could manifest itself as variable performance, especially at peak times (when you might be counting on the service most), higher prices through the back door (to ensure the performance you thought you were getting anyway), or variable pricing (because of the way costs are calculated – e.g. per IOPS).
For a retailer, a ‘normal’ bill of £5,000 a month could rapidly multiply over the busy Christmas period as higher volumes of transactions coincide with peak demand for the service. In one extreme case, photo-sharing startup Everpix went out of business last year after receiving a $35,000 monthly bill from AWS which it couldn’t pay, for storing customers’ photos on its cloud-based storage systems. If consistent high performance and predictable costs are important to you, make sure you find a service that will provide them.
2. Type Of Provisioning
The most common form of cloud-based IaaS relies on hardware virtualisation to ensure that varying tasks can be performed on the same physical machine. But virtualisation carries a performance handicap. The hypervisor uses a part of the machine’s resources and it also introduces delay by interpreting every operation that has to take place on the machine. So it is not as cost-effective as other IaaS services where what you’re paying for and what you’re getting are much clearer.
Shared, virtualised environments do not perform consistently because the loads they are bearing at any given time can vary considerably. When all resources are in play, applications and transactions may run across older, less efficient equipment, driving down performance and driving up cost. Alternative IaaS services such as bare-metal cloud infrastructures work differently, combining the benefits of controlled on-premise infrastructure with the flexible scalability and low maintenance of cloud facilities. Here, the physical server is dedicated to a single user, and there is no hypervisor layer to impede performance.
3. Where Will The Data Reside
Concerns about data safety and the location of data have increased significantly since Snowden and the Patriot Act, so look for a provider that’s completely transparent about its equipment and data locations, about protection against unauthorised access and how content is kept separate from data from other companies. Large-scale commodity services may send data to overseas data centres, which could transgress data safeguarding regulations and otherwise render content vulnerable.
AWS, Azure and Google don’t actually have core infrastructure in the UK. The standard practice is to keep everything destined for the UK in Ireland, and because they are American companies they are subject to US rather than European or UK laws. You will need to choose a provider that is subject to a legal framework you understand, so you can fully comprehend any risks you may be taking with your data and the possible implications for your organisation’s regulatory compliance.
4. What’s The Migration Path?
If you’ve spent years fine-tuning legacy systems that still work perfectly well, you would be right to hesitate about moving everything to the cloud. If you do bite the bullet however, it’s important to make sure you aren’t locking your organisation in to a new environment that will be just as hard to exit if needed in future. If the thought of doing any initial recoding is daunting, the prospect of having to do this all over again in 2-5 years may finish you off.
In many clouds, the use of platform services can be a quick way to get started, but an even faster way to get locked in. Pulling usable data out of proprietary services will not only be a technical challenge, it could also involve significant costs, as providers can charge for network traffic, number of reads/writes or other operations.
5. Is The Service ‘Big Data’ Friendly?
Most businesses are now keen to extract greater value from their data, in the form of new insights that will help them improve efficiency and give customers more of what they want. The cloud brings large-scale business analytics to even the smallest of companies. It might be one the main reasons you’re considering IaaS now. But again not all cloud platforms are equal, as cloud providers are finding as they clamour to offer Hadoop-based data analytics services to businesses on a pay-as-you-go basis.
Apache Hadoop is an open-source software framework for storage and large-scale processing of data sets on clusters of commodity hardware. Its aim is to bring supercomputer-style capabilities to everyday business data. The catch is that the software is designed to run on dedicated physical machines; in a virtualised scenario (as seen in most cloud IaaS environments) performance can slow right down, and bills can soar, leaving companies frustrated and badly out of pocket. Big-data analytics demand reliable, high-speed performance and transparent costs. This is where bare-metal cloud services can offer an advantage, providing a dedicated high-performance server environment but in a cloud-like model because it’s accessed on demand via the web.