Original URL: http://www.theregister.co.uk/2010/11/04/cloud_workloads/
Public Cloud: Sorting out the peaks from the troughs
Baby steps are what you take
Workshop We know many of you remain sceptical of the whole ‘Cloud’ thing and are unsure of what it is, because World+dog seems to be using or abusing the term to peddle their version of a new-age IT vision.
Much of the emphasis on cloud computing has switched recently to focus on the internal IT infrastructure – the so-called “Private Cloud” – because interest in externally provided services – the “Public Cloud” - has been lacklustre to date.
This is a welcome return to reality: cloud computing is not a revolution that will cause us to throw away existing investments and skills and move workloads en masse out of existing data centres to third parties. Instead, “cloud”, in whatever form, private or public, should be seen as a way to incrementally improve how IT is run.
But what to think of the public cloud? Given the low uptake so far a wholesale move to the public cloud is probably out of the question for the foreseeable future. Looking at it on a smaller scale, are there areas where public cloud resources may help with keeping internal servers and infrastructure in peak condition?
Many companies already make use of externally hosted applications or services, for outsourced business functions such as HR or payroll systems, or business applications such as SalesForce.com. The net effect is that these applications or services place less of a load – if any - on the internal IT infrastructure, as the applications run externally. They are in the main, someone else’s problem and their suppliers are paid for it to be this way.
The same approach can be taken when looking to move some workloads out of the internal IT infrastructure and running them ‘on-demand’ in the public cloud. Typically, this is achieved by running the workloads in infrastructure provided by a service provider in an ‘elastic’ manner where capacity is paid for on immediate use, without fixed capacity or a longer term contract.
Capacity and performance can be sourced as desired – within reason – without worrying about commitments. The benefits can be a simplification of the overall scale and complexity of the network and server estate, along with the ability to handle unpredictable workloads without building in too much extra capacity or affecting more critical applications where resources are limited. The key is choosing the appropriate workloads to consider as candidates to run on external cloud platforms.
The IT infrastructure is generally composed of a mix of workloads such as business applications, web servers, databases and communications servers. The workloads may be in production, staging or in a state of development and testing.
These will likely consume resources in different ways. Some workloads may be fairly lightweight, in that they do not tend to use much CPU or memory, generate little I/O or have lightweight demands of storage. Others may be pretty compute-intensive and some may swing between the different states. For workloads that vary, patterns of usage need to be factored in – some may ramp up on predictable days or weeks of the month, while others may peak and trough at unpredictable times.
Public cloud resources can help in several areas.
Using these to provision test and development environments is a leading use case, particularly as these workloads do not require long-term hardware and capacity, and can easily be separated from the production setup.
On the production side of the house, the biggest advantage of public cloud is seen when coping with workloads that may put sudden demands on the infrastructure – these are the ‘bursty’ or periodic types. They may be very demanding on resources during the limited times that they run at full tilt, but in general their overall impact on the IT infrastructure is far less resource-intensive. Compared with over-provisioning infrastructure to cope with the peaks, cloud services may be an attractive alternative.
Compute-intensive workloads are also sometimes seen as good candidates for running in the cloud. These applications tend to make good utilisation of servers and infrastructure, and so may seem odd bedfellows for cloud. However, workloads running with high resource demands can potentially affect other, high-value or high-risk workloads. The case for running them externally may make sense in terms of avoiding impacts on critical business systems.
Looking to the future, cloud offerings will undoubtedly improve with time, making them more and more attractive. At the same time, internal IT should become more dynamic as virtualisation and integrated and automated IT service management become more pervasive, leading to the private cloud.
This will broaden the potential relevance of public cloud services to more mainstream applications and workloads, as the public and private clouds converge. But for the meantime it’s best to evaluate each case on its merits while keeping a firm grip on pragmatism.