Compute architecture
| HPCPERF01: How do you select the compute environment for your application? |
|---|
The optimal compute solution for a particular HPC architecture depends on the workload deployment method, degree of automation, usage patterns, and configuration. Different compute solutions may be chosen for each step of a process. Selecting the wrong compute solutions for an architecture can cause lower performance efficiency.
HPCPERF01-BP01 Evaluate containers or serverless functions
When evaluating compute options, consider containers or serverless functions for your HPC workload or for parts of your surrounding workflow.
Implementation guidance
-
Containers are a method of operating system virtualization that is attractive for many HPC workloads, particularly if the applications have already been containerized. AWS services such as AWS Batch, Amazon Elastic Container Service (Amazon ECS), and Amazon Elastic Kubernetes Service (Amazon EKS) help deploy containerized applications.
-
Serverless functions abstract the execution environment. You can use AWS Lambda to run code without deploying, running, or maintaining, an instance. Many AWS services emit events based on activity inside the service, and often a Lambda function can be initiated off of service events. For example, a Lambda function can be run after an object is uploaded to Amazon S3. Many HPC users use Lambda to automatically run code as part of their workflow.
Key AWS services
Resources
| HPCPERF02: How do you select your computing instances? |
|---|
EC2 instances are virtualized servers and come in different families and sizes to offer a wide variety of capabilities. Some instance families target specific workloads, for example, compute, memory, or GPU intensive workloads. Other instances are general purpose.
Both the targeted-workload and general-purpose instance families are useful for HPC applications. Instances of particular interest to HPC include the HPC Optimized family, the Compute Optimized family and Accelerated Computing instance types that are powered by GPUs and FPGAs.
HPCPERF02-BP01 Select the best computing instance type for your workload by measuring application performance
Select the optimal Amazon EC2 instance type for your workload and consider factors, such as family and generation, to optimize for your desired price-for-performance. With access to on-demand instances, testing different configurations is the best way to determine your desired configuration for each of your workloads.
Implementation guidance
EC2 Instances are available in different generations. Previous generation instances are still fully supported, but we recommend you to use the Amazon EC2 instance types to get the best performance.
Some instance families provide variants within the family for
additional capabilities. For example, an instance family may
have a variant with local storage, greater networking
capabilities, or a different processor. These variants can be
viewed in the
Amazon EC2 Instance types
Within each instance family, one or more instance sizes allow vertical scaling of resources. Some applications require a larger instance type (for example, 48xlarge) while others run on smaller types (for example, 2xlarge) depending on the number or processes supported by the application.
For tightly coupled workloads, optimum performance is obtained when the memory of the computing node is not shared between different virtual machines within the same physical host. Therefore, it is recommended to use only EC2 instances whose size is big enough to occupy the entire physical node. This is usually obtained with the largest instance type even if there are some noticeable exceptions. For example, in the 7th generation of HPC instances, each size in the instance family has the same engineering specs, memory access and price, and differs only by the number of cores offered. That means that all the cores in the instances will be able to access the entire host memory regardless of the instance size. So, you can select also a smaller size without warring about the performance impact of sharing the host memory with other virtual machines.
The T-series instance family is designed for applications with moderate CPU usage that can benefit from bursting beyond a baseline level of CPU performance. Most HPC applications are compute-intensive and suffer a performance decline with the T-series instance family.
Applications vary in their requirements (for example, desired cores, processor speed, memory requirements, storage needs, and networking specifications). When selecting an instance family and type, begin with the specific needs of the application. You can also split a specific workflow in its individual steps (for example, mesher and solver in a CFD simulation) and run each step on a different instance type. Instance types can be mixed and matched for applications requiring targeted instances for specific application components. You can use the AWS Management Console or the AWC CLI to search for instances that satisfy your needs.
As an example, you can use the following command to display only current generation instance types with 64 GiB (65536 MiB) of memory:
AWS ec2 describe-instance-types --filters "Name=current-generation,Values=true" "Name=memory-info.size-in-mib,Values=65536" --query "InstanceTypes[*].[InstanceType]" --output text | sort
Testing different instance types is affordable since you only pay for active usage. Even after your initial choice, you can switch instance types whenever your requirements shift.
HPCPERF02-BP02 Default to virtualized over bare-metal instances
Virtualized instances have a faster initialization time and offer indistinguishable performance when compared to bare-metal instances. Unless you specifically require a bare-metal instance, we recommend virtualized instances, especially in dynamic HPC environments.
Implementation guidance
New-generation EC2 instances run on the AWS Nitro System. The Nitro System delivers practically all of the compute and memory resources of the host hardware to your instances resulting in better overall performance. Dedicated Nitro Cards enable high-speed networking, high-speed EBS, and I/O acceleration without having to hold back host resources for management software.
The Nitro Hypervisor is a lightweight hypervisor that manages memory and CPU allocation and delivers performance that is indistinguishable from bare metal. The Nitro System also makes bare metal instances available to run without the Nitro Hypervisor. Launching a bare metal instance boots the underlying server, which includes verifying all hardware and firmware components. This means it can take longer before the bare metal instance becomes available to start your workload, as compared to a virtualized instance. The additional initialization time must be considered when operating in a dynamic HPC environment where resources launch and terminate based on demand.
Unless your application specifically requires a bare metal node, we recommend using virtualized instances to avoid the longer boot time with metal instances without a gain in performance.
Key AWS services
Resources
| HPCPERF03: How do you optimize the compute environment? |
|---|
You can optimize your compute environment through multiple components, including the operating system and hardware features. Since running in the cloud provides flexibility, we recommend testing different configurations before determining your final implementation.
HPCPERF03-BP01 Optimize your compute environment for your workload
We recommend optimizing your machine image, application-compile options, instance configuration, and runtime environment when running your HPC applications.
Implementation guidance
-
A current operating system running a modern kernel is critical to achieve the best performance and ensuring access to the most up-to-date libraries. An Amazon Machine Image (AMI) is a template that contains the software configuration (operating system, libraries, and applications) required to launch your instance. You can select an AMI with the latest version of the operating system supported by your application. For MPI workloads, it is also important to use a modern MPI version.
-
In addition to choosing an AMI, you can further optimize your environment by taking advantage of the hardware features of the underlying processors. There are three primary methods to consider when optimizing the underlying hardware:
-
Advanced processor features
-
Simultaneous multithreading (SMT)
-
Processor affinity
HPC applications can benefit from these advanced processor features (for example, Advanced Vector Extensions) and can increase their calculation speeds by compiling the software for the target CPU architecture. The compiler options for architecture-specific instructions vary by compiler (check the usage guide for your compiler).
AWS enables Simultaneous multithreading (SMT), commonly referred to as Hyper-Threading, by default on most of the EC2 instances. Multithreading improves performance for some applications by allowing two threads to run on the same physical core. This command will give you the list of the EC2 instances that are offered in a location (or Region) with two threads per core:
AWS ec2 describe-instance-types --filters "Name=current-generation,Values=true" "Name=vcpu-info.default-threads-per-core,Values=2" --query "InstanceTypes[*].[InstanceType]" --output text --region us-east-2 | sort
Most HPC applications benefit from disabling multithreading, and therefore, it tends to be the preferred environment for HPC applications. Multithreading is easily disabled in Amazon EC2 by Specify CPU options for an Amazon EC2 instance for your instance. Unless an application has been tested with multithreading enabled, it is recommended that multithreading be disabled and that processes are launched and individually pinned to cores when running HPC applications. CPU or processor affinity allows process pinning to easily happen.
Processor affinity can be controlled in a variety of ways. For example, it can be configured at the operating system level (available in both Windows and Linux), set as a compiler flag within the threading library, or specified as an MPI flag during execution. The chosen method of controlling processor affinity depends on your workload and application.
There are many compute options available to optimize a compute environment. Cloud deployment allows experimentation on every level from operating system to instance type, to bare-metal deployments. Because clusters are tuned before deployment, time spent experimenting with cloud-based clusters is vital to achieving the desired performance.
Key AWS services
Resources
