What is Performance Computing?
High Performance Computing is an aggregation of computing power to solve problems which are either too large for standard computers, or would take too long.
It’s also referred to as Super computing, For this reason, it enables the simulation or analysis of huge volumes of data that will otherwise not be possible with standard computers.
Challenge of Performance Computing?
The common challenge with high performance computing is that you’re producing or processing more data than your infrastructure resources can handle.
And you have to wait for weeks or months to see the results, which typically delays the innovation and slows down research.
How does it work?
You can think of a high performance computing system as a group of computers called cluster and each computer in a cluster is called a node.
Each unit in the cluster has an operating system, a processor with multiple cores storage and networking capabilities for the units to talk to each other.
A small cluster, for example, can have 16 nodes with 64 cores, which is four cores per processor, helping you solve a problem much faster.
a supercomputer is a much larger variation of this HPC job that would run across an on premises cluster for three months could run across 125,000 cores in 16 hours in the cloud at little or no incremental cost.
And by augmenting your HPC environment with Google Cloud, you gain economies of scale with access to the largest compute and storage hardware, global presence, robust networking, and intelligent automated management capabilities on the cleanest cloud.
How to Build HPC Environment
Now how to build HPC environment on Google Cloud compute, storage and networking other building blocks of high performance computing Compute Engine provides customizable virtual machines running in Google Cloud,
you can scale up and down as needed and choose from a range of machine types for your workloads.
We recommend the compute optimized c2 machines for most high performance computing applications. But you can also select the general purpose and one and two or in 2d for applications requiring larger memory sizes, or a two instances if you need to use GPUs.
for very specific requirements, you can choose custom machine types with the exact number of fours and memory to match your workload requirements and get the best performance per dollar.
You could also use cost effective preemptable VMs, which are short lived compute instances suitable for batch jobs and fault tolerant workloads.
Now the underlying storage system is critical to the performance of many HPC applications. You have a lot of storage options in Google Cloud.
Cloud Storage is a highly scalable object store to store any amount of data. Persistent Disk is durable and high performance block storage for your VM instances.
File store high scale is a high performance scale out file system that makes it easy to mount file shares on Compute Engine. VMs networking,
Google’s privately managed global network infrastructure ensures that your data and applications are the least exposed to the public Internet.
What your private cloud VPC networks are available to enable connectivity from your Compute Engine VM instances, and configure firewalls for your applications. With placement policies, you can control the placement of your VMs in our data centers.
Contract placement policy provides low latency between nodes by placing instances close together within the same network infrastructure to speed communications between notes.
Now, let’s put it all together for your HPC workload on Google Cloud. First, determine the compute storage and networking requirements for your HPC code.
Then create your HPC cluster using Compute Engine instances connected to the storage of your choice.
Additionally, Google Cloud supports several jobs schedulers making it easy for you to start auto scale virtual machines to meet job requirements or spin down a cluster when a job is complete, helping save cost.
Then visualize your result in BigQuery, or AI platform for post processing. From there, just monitor the performance and change the cluster as needed.
Security
Security is important for any HPC workload. Google clouds secure by design infrastructure protects your data applications and users with advanced anti malware and threat detection.
High Performance Computing drives research, development and innovation in many industries from rendering the visual effects in the latest blockbuster movie.
sequencing the human genome to cure diseases, to risk analysis and financial services or even to design the next generation of cars.
