Running PRIMO in the Amazon.com elastic compute cloud

Without investing in expensive hardware upfront, windows virtual machines can be launched, configured and connected using the Amazon Elastic Compute Cloud (EC2) where PRIMO simulations can be deployed in mere minutes.

Amazon EC2 is the central core of Amazon.com’s on-demand cloud-computing platform. It provides scalable computing capacity where one can launch as many or as few virtual servers as one needs, configure security and networking, and manage storage through a web-based user interface. The EC2 virtual computers also called instances, come with preconfigured templates known as Amazon Machine Images (AMI) which contain the operating system and other software [6]. This allows a user to install and run specific applications such as PRIMO in just a few minutes via a client computer using the remote desktop protocol. One key difference however between a real server and an Amazon EC2 server is that when an instance is terminated, the virtual server and its data are no longer available.

Access to Amazon EC2 is on a subscription basis and one is charged for the usage on a per-hour basis. There is a free-tier option for new account holders for the first twelve months of use where there are no charges incurred. The free-tier option allows a new user to run a micro-sized (1 CPU, 1 GB memory) server in the cloud, with storage and bandwidth completely free of charges for one year, provided the monthly usage does not exceed 720 h.

The PRIMO program of the same version is first installed and run in a Windows 2016 base server AMI with instance type in the free-tier option. The simulation of a 6 MV Varian C2100 linac with 6.26 MeV initial energy is configured to run for 720 h. After a full 30 days of non-stop calculations, the s1 segment is completed with a total of more than 1.07 × 10⁷ histories.

In order to simulate huge number of histories in a much shorter period, a faster EC2 instance is launched – the C5 instance, introduced in 2017, powered by 3.0 GHz Intel Xeon scalable processors. With this type of architecture, a Windows 2019 base server AMI with 32 cores and 64 GB memory takes more than 53 h to complete the s1 segment of more than 2.75 × 10⁸ histories and another 7 h to finish the s2 and s3 segments in the binned water phantom for the tuned beam 10 MV configuration.

The simulations in the Amazon cloud including the bandwidth for file transfers to the local computer are totally free of any charges because of the free-tier option and other educational credits offered by Amazon Web Services (AWS) [7]. Remarkable as that may seem, what is really phenomenal is the fact that even without these educational credits, the charges for the full simulation of the tuned beam 10 MV configuration amount to just around two hundred dollars (US$200) at 2019 prices. Nevertheless since the educational credits available are not unlimited, the use of AWS EC2 platform may no longer be recommended for the fine-tuning of the beam profiles since the entire procedure involves many stages; unless of course the financial resources are not an issue.