High-performance computing (HPC) is used four industries such as genomics, finance and financial risk modeling, machine learning, weather prediction, and even autonomous driving.
Data transfer
Snowball, Snowball Edge, Snowmobile
Snowball/Snowmobile is the a big data transfer device in which you load terabytes/petabytes worth of data and then ship off to AWS. AWS will uses their backbone network to load data in it to your S3 bucket.
Snowball Edge is a 100TB data transfer device with on-board storage and compute capabilities, which means you can embedded independent programs such as Lambda functions in it. Snowball Edge connects to your existing applications and infrastructure using standard storage interfaces. Snowball Edge can cluster together to form a local storage tier and process your data on-premises, helping ensure your applications continue to run even when they are not able to access the cloud.
Once the data transfer job has been processed and verified, AWS performs a software erasure of the Snowball appliance.
DataSync
DataSync is an agent on a virtual machine that you can install on your on-premise data host. Then you can use it to upload your data to S3, EFS, FSx, etc.
Direct Connect
AWS Direct Connect helps your establish private connectivity between AWS and your data center, which reduce your network costs, increase bandwidth throughput, and provide a more consistent network experience than internet-based connections.
Compute and networking
EC2 instances that are GPU or CPU optimized
EC2 fleets (Spot instances or Spot Fleets)
Placement groups (Cluster placement groups)
Placement groups can be used to decrease latency between communication of EC2s.
Enhanced Networking
It uses single root I/O virtualization (SR-IOV) to provide high-performance networking capabilities on supported instance types.
Compared to transitional virtualized network interfaces, SR-IOV has higher I/O performance and lower CPU utilization.
Enhanced networking provides higher bandwidth, higher packet per second (PPS) performance, and consistently lower inter-instance latencies.
No additional charge.
Depending on your instance type, enhance networking can be enabled using an:
- Elastic Network Adaptors (ENA) with speeds of up to 100Gbps
- Intel 82599 Virtual Function (VF) interface with speeds of up to 10Gbps. This is a legacy product and is typically used on older instances.
Elastic Fabric Adaptors (EFA)
Elastic Fabric Adaptors (EFA) is a network device you can attach to your Amazon EC2 instance to accelerate HPC and machine learning applications.
It uses OS-bypass, which enables HPC and machine learning applications to bypass the operating system kernel and communicate with the EFA devices directly.
Currently not support on Windows.
Storage
Instance-attached storage
- EBS with Provisioned IOPS
- Instance Store
Network Storage
- Amazon S3
- Amazon EFS
- Amazon FSx for Lustre
Orchestration and automation
AWS Batch
- Supports multi-node parallel jobs, which allows you to run a single job that spans multiple EC2 instances.
- Enables to run thousands of batch computing jobs on AWS easily.
- Schedule jobs and launch EC2 according to your needs.
AWS ParalllelCluster
- An open-source cluster management tool that makes it easy to deploy and manage HPC clusters on AWS.
- Use a simple text file to model and provision all the resources needed for your HPC applications in an automated and secure manner.
- Automate creation of VPC subnet, cluster type, and instance types.