EC2 Dashboard

Amazon Elastic Compute Cloud (EC2) is a service that provides scalable and flexible computing capacity in the AWS cloud. Using Amazon EC2 eliminates the need to invest in hardware and enables quick development and deployment of applications. It allows many virtual servers to be configured with security and networking, and storage management. EC2 virtual servers, also known as instances, are the building blocks for supercomputing on AWS.

The EC2 dashboard contains resources and the ability to launch an instance. On the left-hand side of the dashboard there are many links, including to EC2 limits, instances, AMIs, Security Groups and SSH keys.

ec2-1

Instance Types

AWS has nearly 400 instance types, which are organised using a standard naming convention. You can see the full list of these instances in your AWS console page or via the EC2 website here.

However lets take the example of the c5n.18xlarge instance, which will be used in this workshop.

  • The ‘C’ stands for compute optimized, which have a memory (RAM) to CPU ratio of 2-5 (as opposed to M or R instances which have higher RAM to CPU ratios).
  • The ‘5’ stands for the fifth generation of this particular instance (there is a C4 instance which is based upon an older generation of the Intel chipset).
  • The ‘n’ standard for network optimized, which means it supports the 100 Gbit/s Elastic Fabric Adapter (EFA) adapter. This is needed to achieve the strong scaling to allow your code to run on thousands of cores. You can read more about EFA here.
  • The ‘18xlarge’ for this instance is the largest virtual machine option for this instance. This means you are the only user of this particular instance (server) and can obtain the maximum network, memory and processors that it contains.
  • For multi-node CFD runs, it is always better to pick the full instance size. The number will change depending on the chipset design but for the c5n, it’s 18 because there are 18 cores per socket, with two sockets making 36 physical cores in total. Finally you should be aware that any Intel/AMD chip is quoted as vCPU’s e.g 72 in the case of the c5n.18xlarge, however for CFD we typically advise to run without hyperthreading, therefore there are 36 physical cores available.

For CFD we typically recommend c5n.18xlarge (see benchmark section) unless there is a particularly large mesh where a serial part e.g the surface wrapping may need more than 192GB RAM and there you may want to run the mesh portion on a M or R instance (which have x2 or x4 higher memory) e.g the m5.24xlarge, which has 384GB RAM and 48 physical cores.

If you want to hear more about EC2, then please watch the video below.