High Performance Computing (HPC) is in use across Connecticut. Our state is home to a variety of data intensive industries that rely on HPC systems. Though it is possible to host HPC instances in the Public Cloud, there are benefits to locating these powerful systems at colocation data centers.
High Performance Computing, as the name suggests, employs speedy multi-core CPUs or GPUs (graphical processing units) along with fast storage systems and memory to process information quickly. The market for HPC systems is growing as the demands of Artificial Intelligence, Image Processing, and large data base applications surges.
Connecticut is home to industries that are investing in HPC. Higher education leaders such as Yale and the University of Connecticut have ongoing HPC programs in place. In-state life sciences companies including those in the biopharmaceutical, medical imaging, and health research industries are taking advantage of these powerful computer systems. Our state is also home to many insurance and finance companies that require these powerful systems. Aerospace, defense contractors, and high-end manufacturers located in our state also are increasing their use of HPC systems.
Colocation Costs Less Than Public Cloud For HPC
Colocation is a better choice than the Public Cloud for HPC applications for several reasons. First, colocation is often much less expensive. Whereas it is possible to configure instances with GPUs and high-performance memory and storage, the Public Cloud has not been optimized for these types of uses.
A recent article “Colocation vs Cloud: SEO Firm Finds Cloud to be Cost Prohibitive for its Clusters of Powerful High-Density Computers” (Network World 4/4/23) documented the cost disparity.
The article describes how a search engine optimization SaaS company estimated it would cost an additional $400 million over 3 years to operate in the Public Cloud. The cost of procuring HPC infrastructure in the AWS Public Cloud was compared to purchasing 850 HPC Dell servers and hosting them at a colocation facility.
Avoid Public Cloud Latency
Latency is also an important consideration. Most Public Cloud data centers are located hundreds of miles from Connecticut. For example, AWS’s Northern Virginia data center is over 300 miles from most locations in Connecticut. Data sent over the internet typically experiences, on average, an .82 ms delay for every 100 miles traversed. So, the additional roundtrip latency imposed by the Public Cloud could be over 4.8 ms versus a colocation facility in Connecticut. When organizations are investing a lot of money to achieve optimal performance it does not make sense to slow things down, even by a few milliseconds, with the latency burden of the Public Cloud.
Much like a schoolteacher can pick out brighter students just by observing them, it is often possible to identify a HPC system when walking through a data center. High Performance Computer systems generate more heat output since they consume more electricity to run their powerful servers. Also, often they are noisier since their internal fans are running a lot to disperse heat.
Fortunately, CAPS has plenty of available floor space, power, and cooling capacity to handle the growing High Performance Computing needs of Connecticut and the Northeast region. Our data center in Shelton is centrally located to minimize latency. Please contact us to discuss your HPC hosting requirements.