Given its position in the technology sector as a leading chipmaker, Intel is perhaps not a typical business when it comes to internal IT. But it experiences the same challenges faced by many large businesses.
IT director Chris Shaw (pictured) has responsibility for the company’s IT across Europe, the Middle East and Africa. Like many large companies, Intel is moving forward with virtualisation and cloud computing, he says.
Intel’s cloud journey started about two years ago. Any IT organisation may have a huge amount of compute power, but the challenge, according to Shaw, is knowing what is available for a new application.
A few years ago, prior to the server virtualisation project, Intel had 20,000 servers with average utilisation of only 38%, says Shaw.
Like many IT department heads, he faces difficult conversations with business managers who want their own servers, even when there is a compute cloud available that provides processing on-demand for most application requirements.
“IT in Intel is very much like IT in any organisation, where some teams will fight to keep dedicated resources for their applications," says Shaw.
He says the IT department must provide the data to prove it can save money and explain to each business manager that they can have a unique instance, full autonomy of what is installed, and that the IT department will ensure performance is not undermined.
More CIO interviews from Computer Weekly
Virtualisation across the board
The IT team at Intel defined ratios for physical servers to virtual machines (VMs) by monitoring use, and then negotiated with business managers to calculate an 18-month projection of anticipated growth, he says.
One of the benefits of modern chips is that the hardware can be segmented, allowing applications to run on physical processor cores, says Shaw.
While it is possible to dedicate specific cores for VMs, it is not something he would recommend, since dedicating VMs to specific processor cores is not that different to dedicating hardware to applications.
“We have moved from a physical server infrastructure to a setup where almost every application is virtualised,” adds Shaw.
It has taken almost three years to become fully virtualised. Intel began with small capabilities. "Web applications are low risk and give us the ability to see how [the infrastructure] can scale,” he says.
Intel has since advanced to virtualising enterprise systems. “We had a significant ERP [enterprise resource planning] upgrade and did software and hardware migration, which gave us economies of scale,” he says.
One of the benefits of running enterprise systems on a virtualised infrastructure is that databases can be load-balanced across VMs, which makes them more scalable, and the VM offers built-in redundancy, according to Shaw.
“In practice, our current implementations of database load-balancing across VMs offer superior performance compared with using the RDBMS [relational database management system] 'clustering' capabilities,” he says.
Even applications such as simulations, which one would assume need dedicated hardware, are now run on Intel’s virtualised infrastructure. “In chip design we run simulations. We had assumed we would not get much improvement, but thanks to hyper-threading, we found we were getting a 20% boost in performance and it is cheaper than physical processor cores,” says Shaw.
With hyper-threading, the operating system treats multiple processor cores as if they were distinct processors, which boosts performance.
Shaw says this performance boost has other benefits within Intel: “We can accelerate video in YouTube, which means we don’t need quad-core machines, when a dual-core system will do.”
Custom-made virtualisation management software
Rather than use off-the-shelf management tools for virtualisation, Intel created custom software to automate the provisioning of VMs. Its in-house software supports features such as self-provisioning.
“Any team can build out hardware infrastructure and provision VMs, which are created from VM templates,” says Shaw.
Download additional resources
Intel built a monitoring environment that looks at VM sprawl, where virtual machines are created but never removed from the infrastructure, so they continue to consume resources. Killing these unused VMs is a bit of a judgement call for the IT team, he adds.
Shaw says the virtualised infrastructure has increased server utilisation by 17%: “We increased compute cloud utilisation from 38% to 55%, which enabled the company to save $5.6m.”