The increased performance required by high performance computing is obtained through the use of parallelism. Three primary components are required to create an HPC system. Computing power is provided by networking many compute servers. These servers are known as nodes and allow programs and algorithms to be run simultaneously on each node to increase the speed at which calculations are performed. A fast network and storage infrastructure is required to keep up with the processing speed of the nodes.
At one time, the only means of attaining HPC was through the use of large, mainframe supercomputers. These machines are prohibitively expensive and beyond the means of most organizations. A more economical alternative is to create clusters of nodes from inexpensive components, but this method is still beyond the financial capabilities of many enterprises that may have limited uses for HPC capabilities.
HPC Cloud Offerings
Public cloud providers have developed HPC functionality that can be used on an as-needed basis. Lowering the costs of high performance computing promises to bring its benefits to many more organizations that could not justify the expense of an on-premises implementation. A company can use cloud-based HPC to conduct research and run simulations that were previously out of reach which allows them to compete with much larger rivals.
Major cloud providers have added HPC to their portfolios and are leveling the playing field for organizations of all sizes. The following are some examples of available cloud HPC offerings.
AWS HPC - Amazon Web Services offers HPC functionality that includes an integrated suite of services designed to enable enterprises to quickly build and manage HPC clusters. Customers can make use of artificial intelligence and machine learning functionality, fostering innovative solutions.
Azure HPC - Microsoft provides users with a set of computing, networking, and storage resources optimized for high-performance workloads that reduce job execution time from days to minutes.
Google Cloud HPC - Using Google resources, organizations can essentially build a custom supercomputer while retaining the flexibility to only pay for computing power that they need.
These are just a few of the cloud HPC offerings that are bringing enhanced computing power to anyone who needs it.
Overcoming Obstacles to Implementing HPC in the Cloud
Some potential issues may influence whether an organization wants to take advantage of cloud HPC. In cases where an enterprise needs to make extensive use of HPC capabilities, they may opt to retain their on-premises systems. A few considerations when engaging a cloud HPC solution include:
- Data Security - Security is a priority for all organizations and in the cloud it is a shared responsibility between the customer and provider. In some cases, an enterprise may want more control over data security than possible in the cloud.
- Data transfer - It can be challenging to move the large quantities of data used in on-premises HPC systems to the cloud.
- Performance - Latency in cloud networks may be of some concern to customers interested in cloud HPC.
Cloud providers are addressing these issues as cloud solutions continue to mature and expand their capabilities. It seems that the benefits of inexpensive HPC systems may outweigh the potential problems for all but a few prospective customers.