Companies wasting £1 billion a year on underused Cloud capacity

Cloud server provider, ElasticHosts, is announcing that Linux IaaS customers are being unnecessarily overcharged a total of £1 billion each year, based on an analysis of independent industry figures. The prevalent model for IaaS is based on the use of virtual machines, where compute is paid for based on what is provisioned, but this is at best 50% utilised by typical workloads. However, recent updates to the Linux kernel mean containerised servers can now be offered to cloud customers, where they are only charged for capacity they actually use, rather than what they provision, and as a result £1bn of spend on wasted IaaS capacity could be eliminated.

£1 billion wasted – show me the money!
Gartner estimates that the global IaaS market is worth £5.4 billion. Of this market, Forrester claims that 47% of global web developers (representing £2.538 billion) use IaaS for compute, rather than storage or print; Frost & Sullivan claims 73% of the US IaaS market are compute users (£3.942 billion). As such, by taking a conservative average of 50%, based on the global figure from Gartner, the overall market for IaaS compute users is at least £2.7 billion.

Breaking this down further, the Linux Foundation estimates that 76% of organisations are using Linux servers for cloud, meaning Linux equates to £2.05 billion globally of the IaaS compute market. Figures from IBM, McKinsey, Google and Gartner estimate that servers are currently utilised anywhere from 6% to 40% on average; we can therefore safely estimate that IaaS is being underutilised by a minimum of 50%, resulting in at least £1 billion (£1.026 billion) of wasted spending on Linux based cloud compute each year.

Richard Davies, CEO and co-founder for ElasticHosts, had the following to say: “The latest updates to the Linux kernel have made containerised IaaS possible. This has huge ramifications for the IaaS market as customers no longer need to over provision – they are only charged for what they actually use and can therefore cut costs in half. In the same way that virtualisation disrupted the physical server market; containerisation is set to disrupt the cloud market and has the potential to deliver £1 billion worth of savings to Linux cloud compute users every year.”

User profiles – where is the waste coming from?
ElasticHosts looked into specific use cases from its own customer data and found server under-utilisation was a problem for many different users, across a wide range of usage patterns. The three most common use cases that stand to benefit from the technology are Web Applications, Testing and Development and Disaster Recovery.

Web Application Users
Web application customers tend to run cloud servers on a 24/7 basis, but utilisation patterns vary significantly throughout the day with web traffic. The graph below shows a typical web application running on a fixed cloud server set at 4GB RAM. As you can see, there are regular intervals when the server is being underutilised, but the customer is still paying for 4GB of capacity. At the same time, there are periods with a sudden surge in web traffic where there is not enough capacity allocated to the server; this results in performance issues, such as slow page loads or load errors. To avoid this, they would have to constantly monitor performance and provision extra servers as needed; or set capacity at an uneconomically high level. Organisations deploying web applications can typically save around 50% through using auto-scaling containers, as opposed to fixed-size virtual machines, since these servers can scale immediately up to 64GB and back down again, and users are only ever charged on the exact amount of capacity they actually use.

Testing and Development Users
ElasticHosts also looked at customers running servers for testing and development and estimated these customers could also save a minimum of 50% through using auto-scaling containers, rather than fixed-size VMs. As the graph shows, a typical development team will run tests intermittently. Users do have the option of turning the server off when it is not being used, but many work in large development teams and for the sake of convenience they generally leave it turned on ready for the next user. However, this can mean the server can go unused for days, even though they are still being charged for the capacity. Notice the graph also shows that during heavy automated testing, the application exceeds the amount provisioned by the server, again affecting performance and testing results. With containers, users can simply create a container; set an upper limit of capacity; start the test; and forget about it, complete in the knowledge that they will only ever pay for capacity they use.

Disaster Recovery Users
Disaster recovery (DR) is the clearest example of waste. DR customers routinely replicate 50-100% of their servers as idle ‘hot spares’, constantly provisioning them at full capacity. As a result, they are effectively paying for capacity twice. ElasticHosts estimates that the costs of a fully-configured duplicate of the primary server that is running continuously could be reduced by 80% or more, if these idle ‘hot spare’ servers were billed on actual utilisation, rather than the amount provisioned.

Davies concludes, “While usage patterns vary, it is clear from these graphs that most cloud servers frequently run at low utilisation and are not delivering on the early elastic promises of cloud. Advancements in containerisation technology mean these users no longer have to overprovision and sacrifice price for peak performance. The sooner the rest of the cloud industry offers containers, the quicker we will evolve to the next generation of the cloud, and the more customers that will benefit.”