Demand for high performance computing is on the rise. From big data analytics to Bitcoin mining to genomic sequencing, more organizations are running applications that are computationally and/or data intensive.
As I wrote in The Buzz: Use Cases & Trends in High Density IT Infrastructure, when figuring out how to accommodate high performance computing (HPC) applications in the data center, IT organizations have broadly two choices: many typical server racks spread across a large data center space; or fewer specialized (high density) server racks consolidated in a smaller data center space that is set up with power and cooling that can support high density hardware.
But what are the factors that organizations consider when deciding how to accommodate increasing demand for HPC capacity? Here, I share what we’re hearing from organizations facing that decision, and what they’re saying about why they might choose a high density or a low density environment.
1. 5kW or 25kW – how much power will your server racks consume?
For organizations with servers that can run at high densities, when facing the need to increase capacity it can be more efficient to push those higher densities rather than invest in new servers altogether. That was the case, for example, for a financial services firm that needed to add more compute capacity in 30 days or less and wanted to increase rack density with high performance compute to squeeze the most out of their hardware.
2. Do you have space to grow?
Accommodating high performance computing in a low density environment means spreading out over a relatively larger data center footprint than in a high density environment. For example, in a modular data center built to suit high density, consumers can get the same amount of compute in 462 square feet as in 3,000 square feet (or more) of traditional raised-floor data center space. That is a big deal in geographies like New York and London, where there is demand for HPC but real estate is highly constrained (and very expensive).
3. Can your current data center infrastructure support the power and cooling needs of high density hardware?
Not all data centers are equipped to handle the intense power and cooling requirements of high density infrastructure. Forrester explains, “[IT infrastructure and operations] pros have told Forrester on several occasions that converged infrastructure offerings are more power-intensive and require more cooling than their data center is prepared to handle.”
Where high density infrastructure typically supports 20-25kW per rack, “the typical data center can only support an average of 5kW per rack,” according to Data Center Knowledge. In contrast, Data Center Knowledge reports, it is not at all uncommon to see 20-25kW densities in a modular data center. (In this report, 451 Research explains why prefabricated modular data centers suit high densities.)
4. How significant a concern is energy efficiency?
A high density server rack consuming 25kW of power is obviously more power-intensive and cooling-intensive than a low density server rack consuming 5kW of power. But when high performance computing applications are set up to run on a cluster of low density server racks, spread out over a larger data center footprint, the environment actually requires more power and more cooling. Data center expert Mark Thiele explains: “In fact, even the airflow management (ducting, etc.) when done effectively doesn’t need to change significantly regardless of density, it’s the ability to cycle the right amount of air to and from the correct locations that matters.”
5. Are you likely to need additional compute capacity again in the near future?
In 2013, the amount of information processed online – the global “digital universe” – was 4.4 zettabytes (4.4 trillion gigabytes) big. By 2020, it is estimated to be ten times that size. That trend is reflected in increasing demand for high performance computing capacity to do something with all that information, and there’s no indication that the rise will abate.
So another consideration factor is scalability. As the program manager for Lawrence Livermore National Laboratory’s high performance computing facility explains, “We want to get out of doing construction over and over, and just build what we need when we need it…so as the HPC technology changes, we can adapt.” (For that HPC facility, modular was the model of choice to “future proof” the data center.)
Learn MoreDensity Calculator – Modular vs. Raised FloorThe Buzz: Use Cases & Trends in High Density IT Infrastructure blog post