The team started in the beginning with companies such as BMW, Volkswagen, financial institutions, and research institutions, such as the Earlham Institute. They have their hardware and their own equipment in Iceland in the Verne Global secure campus delivering high-performance computing output. What Verne Global found was, in order to really bring the small and medium businesses forward into HPC - it is expected that the future innovators really will be leveraging high-performance computing - they are going to need industrial scale, if they want to be successful in leveraging this. Here, Verne Global comes in, Tate Cantrell stated, because the company develops not only data centre campuses but the HPC compute Cloud as well. Verne Global has the financial capabilities to deliver these at a true industrial scale.
Tate Cantrell wanted to tell about how hpcDIRECT works and why it is relevant to the high-performance computing community. First of all, hpcDIRECT is delivered on bare metal. Verne Globla's customers have a one-to-one relationship with their own compute. The company's experience is that customers can't get this within the traditional Cloud. There, they have challenges such as virtual machines meaning that where the networking is, is unknown in between. The company has very high specification machines with Xeon processors - Skylake 6130 and 6154, RAM standard all the way up to a terabyte of RAM, NVMe, and SSD storages. All of the nodes are interconnected with InfiniBand EDR, as well as optional multiple Ethernet networks. This allows Verne Global's customers to be able to compute in a Cloud environment, both reserved or in an on-demand burst model. It allows them to compute with confidence and be able to port even their most complex MPI algorithms into Verne Global's Cloud and operate.
Since the company uses OpenStack as the basis for its orchestration layer of the bare metal computing, the team can create a dedicated blueprint for each of Verne Global's customers. They can use industry standard tools such as Terraform, a tool set under the Mozilla public license, to be able to build that stack the way they would like. The company uses elements which are simply answerable roles to deploy the specific libraries and software applications that a particular customer requires. Once that blueprint is built, it is both portable to other Clouds for federated purposes, but it is also a push-button environment. If the customers want to grow their environment from 10 nodes up to 100 nodes, that is very easy to do. It is just a variable modification within that blueprint file, explained Tate Cantrell.
Companies like Satavia, an HPC company that is focused on solutions for the airline industry, are able to come into the Verne Global Cloud and have the flexibility to scale all the way up to 10,000 nodes on their solver which is computing the formation of high atmospheric ice clouds that are very problematic to the airline industry, since they can cause flame out during turbine run. What Tate Cantrell likes about companies like Satavia has two reasons. Firstly, they are the only ones that he is aware of in this community, that are working very deeply on legacy data. This obviously increases the amount of computational effort required to make the predictions. Secondly, they have years and years of data but they also have an experience of having worked in the public Cloud environment. As such, they know both the benefits of Cloud but also the shortcomings of public Cloud. When they found that with hpcDIRECT they were able to work directly with their machines and make requests on BIOS modifications to those machines, that gave them the power to be able to deploy their highest priority workloads into a Cloud environment, avoid the capital expenditure (CAPEX) cost and be able to shift that CAPEX into an operational expense model (OPEX).
Primeur magazine: If you look at the customers that you have, can you say that it is spread among SMEs, big companies, and research institutes?
Tate Cantrell:It is spread, yes. Verne Global has large enterprises, financial institutions, and research companies among its customers but the biggest growth opportunities are in the small and medium sized business market. Verne Global believes that this is where over the next decade the majority of growth is going to occur within the HPC market. The company thinks about the SMEs and what they require. They require a higher level of support than the traditional enterprise. If you think about the way the enterprises have deployed high-performance computing, it has been at scale for a number of years. They already have the procurement teams and the operational teams to be able to operate that environment. However, if you have an SME with a really clever idea that's perfectly suited for high-performance computing, sometimes they don't have the DevOps expertise which the Verne Global team has, because this team consists of HPC experts, who are delivering solutions for their customers that are innovating with HPC. Verne Global believes that innovation will come from SMEs. They require HPC but they also require it to be easy to take down and that's what hpcDIRECT provides.
With a location like Iceland, it's not just the 100% renewable energy that Verne Global is able to deliver at scale, which is some of the cheapest power in the European continent but you also get the benefits of free cooling. The company has a number of just truly unique data centre solutions that it has been able to build there that you can't build anywhere else in the world. Tate Cantrell has been working in the data centre industry and actually started in a Cray supercomputer cluster in Louisiana in the United States, back in 1994 or 1995. He has been around for a while. He built a lot of Big Data centres but never has been able to build some of the unique designs that Verne Global has been able to build in Iceland because of the natural free cooling and the access to truly dependable power that is delivered at industrial scale.
Primeur magazine: Sometimes you wonder why not all of Iceland is covered with data centres.
Tate Cantrell:Some people would point out that machines at Verne Global are not next to the eyeballs. From a network standpoint, they are absolutely correct. The company has great network connectivity to both North America and Europe but there is latency involved. That is one of the reasons that high-performance computing is perfect. Verne Global is working with a company in Australia that is doing work for the oil and gas companies. All they require is good output, good input, and good remote visualisation. If you have the right application and the right location, you'll get truly optimized performance. This is what Verne Global is all about, it is supporting the customers providing them performance, providing them support, providing the security and true ownership for their equipment.
Primeur magazine: Thank you very much for this interview.