Overcoming the struggle to take machine learning projects to the cloud
There is no question that machine learning is a hot topic these days, and that the implementation of ML algorithms in numerous industries has shown proven benefits by helping companies automate tasks to offload this work and ultimately help them focus on their core business processes.
As with Linux, virtualization, cloud computing and containers before this, machine learning has also risen to become one of the most in demand skills in 2018. As a result, many organizations that are looking to deploy machine learning are also working with limited resources when it comes to migrating their machine learning and high-performance computing (HPC) workloads to a cloud or hybrid environment.
With this in mind, I wanted to understand what our customers are experiencing when it comes to machine learning projects, so we sponsored a survey to identify the specific types of projects driving value in machine learning, as well as better assess what key challenges users are facing that are preventing them from moving their projects into production.
We were expecting some of the results, but many of them showed valuable trends for us to be aware of when talking with organizations with HPC workloads.
The survey, conducted by Dimensional Research, interviewed 344 technology and IT professionals (representing several stakeholder positions within these groups), across the globe and included 17 industries.
With this group in mind, it was very interesting to see that more than 90 percent of respondents said that are already using machine learning and that ML projects will continue to grow over the next two years, which signifies explosive growth in this market over the next couple of years. Also, though a diverse set of ML projects are currently initiated by 93 percent of the respondents, only 22 percent of these projects have moved into production. This last piece of data made us want to dig into the results a bit further.
So, we looked at how many teams within these organization surveyed were running ML projects and noticed that 69 percent of companies surveyed have three or more teams requesting ML projects. Yet as mentioned, only 2 in 10 companies have ML projects running in production.
So when it came to better understanding what some of the key components driving these user’s ability to successfully move ML projects into production, here is what we uncovered:
- There is a direct correlation between HPC and ML, with more than 88 percent of respondents indicating that they are working with HPC in their jobs.
- Nearly 9 out of 10 companies surveyed expect to use GPUs as part of their ML infrastructure.
- More than 80 percent of respondents plan to use hybrid cloud for ML projects while keepings costs down.
The infographic below summarizes the top statistical information related to users’ perceptions around machine learning projects.
What this information tells us is that while many technology and IT professionals are taking advantage of, or expecting to take advantage, of HPC, GPUs and hybrid cloud for their ML projects, they are challenged with how to migrate these workloads to the cloud.
Most of our customers seeking to deploy machine learning often do so by refactoring or introducing new code or algorithms into either net-new or existing applications. They turn to us, for example, once they’ve had success in R&D prototyping on isolated laptops and workstations and want to discuss how to implement ML in production and across their organization. When they hit this stage, the typical problems they run into are often around scaling, sharing, integration and containerization. Therefore, it is important for users to understand how to leverage hybrid cloud for their ML projects.
Through our conversations and the trends, we see currently taking place in the industry, it’s obvious that we’ve reached a tipping point, and hybrid cloud computing can be an essential approach to HPC architectures to help realize value with machine learning projects.
By using the right strategies, enterprises can maximize local resource usage, reduce total costs and improve productivity and capacity by bursting to the cloud. For HPC users working on machine learning projects, the question is no longer whether to embrace hybrid cloud, but rather, how to start.