CFD Direct is pleased to announce the release of version 8 of CFD Direct From the Cloud™ (CFDDFC®). CFDDFC v8 is available as a pay-as-you-go image for Amazon Elastic Cloud Compute (EC2) on Amazon Web Services (AWS) Marketplace. It is available as: a) the standard CFDDFC product; and, b) CFDDFC (Arm).
CFDDFC is the marketplace product for the majority of instance types, based on x86 processors such as those from Intel and AMD. It includes a full desktop environment with graphical applications, accessible using the X2Go remote desktop (see below). It includes:
- OpenFOAM v8 released by the OpenFOAM Foundation, produced and maintained by CFD Direct;
- ParaView v5.6.0 for data visualization;
- OpenMPI v3.1.4 with support for Elastic Fabric Adapter (EFA) for HPC workloads;
- FreeCAD v0.18.3 parametric 3D CAD modeler;
- … running on Ubuntu 18.04 LTS GNU/Linux operating system.
CFDDFC (Arm) is the marketplace product for instances using processors with Arm architecture, notably AWS’s new C6g instances with their Graviton processor. CFDDFC (Arm) is a server-only distribution without remote desktop and graphical applications. CFDDFC (Arm) includes:
- OpenFOAM v8 released by the OpenFOAM Foundation;
- OpenMPI v4.0.3;
- … running on Ubuntu Server for Arm 20.04 LTS GNU/Linux operating system.
EC2 includes Compute Optimized instances for compute-intensive workloads. The C5 instances have a range of sizes up to
c5.24xlarge which contains 48 physical cores. C5n instances are a variant of C5 that provide significantly higher network performance across all instance sizes, with the largest C5n instance,
c5n.18xlarge, containing 36 physical cores.
The C6g instances with the Graviton2 processors include a range of instances sizes up to
c6g.16xlarge which includes 64 physical cores.The clock speed is lower than the C5 instances, making the performance of
c5n.18xlarge broadly similar. The C6g instances provide up to 40% savings in on-demand running costs for equivalent performance. This makes C6g very attractive for: batch runs, where graphics is not needed; on-demand runs, which cannot exploit the cost savings of spot instances.
Clusters of Instances
CFDDFC v8 provides high performance computing (HPC) by clustering instances with faster networking to deliver good scalability. CFDDFC supports Elastic Network Adapter (ENA) which provides the high throughput, low latency networking required by clusters. Both C5n and C6g instances support ENA, with
c5n.18xlarge having 100 Gbps of network bandwidth compared to 25 Gbps with
A cluster of C5n instances delivers 70%-90% scaling at 504 cores. This performance is significantly better than standard C5 instances, which exhibit scaling below 50% at the same number of cores. Even with the higher price of ~25%, C5n instances are more cost-effective than C5 when clustering multiple instances.
The standard CFDDFC also supports Elastic Fabric Adapter (EFA), an additional network interface for HPC applications running on
c5n.18xlarge instances. CFD Direct previewed the EFA technology, which showed ~70% scaling at 1008 cores for meshes of 100 k cells per core and linear scaling at 1008 cores for a fixed mesh size of 97 m cells.
Command Line Interface
cfddfc command line interface (CLI) manages instances of CFDDFC on AWS, transfers data to and from instances and provides convenient access to the instance by remote desktop or SSH login. It automatically selects the relevant CFDDFC v8 product to suit the instance type selected at launch. For example, the CFDDFC (Arm) product is used with the following launch command for a 64 core
cfddfc launch -instance c6g.16xlarge
A user could then create a cluster of 4 instances, i.e. 256 cores, by adding 3 slave instances using the
cluster subcommand, e.g.
cfddfc cluster -slaves 3
Once the slave instances are running, the cluster is fully operational.