Features, STARTUPS

Supercomputer on demand and supercloud for hire

A year ago Russian startup HPC Hub launched a new service, a supercomputer on demand, offering businesses a short term rent of Rosatom State Atomic Energy Corporation’s supercomputer with its all 1,600 kernels and 64 GB RAM. The idea turned to be beneficial for all. The owners of the expensive hardware could lease their underexploited capacities, the software developers could monetize a new product, the supercloud, while entrepreneurs, mainly medium businesses, got access to a unique IT infrastructure which would otherwise be fabulously expensive for them. This Russian service based on a simple shared economy idea in HPC (High Performance Computing), could be the first in the global market, but hesitancy of its investors did not let it take the lead. Still, HPC Hub is confident its leadership is still possible.

Credit: hpchub.net

Per aspera ad nubes

HPC Hub’s project emerged in 2013 as a research task set by a team of lecturers and graduates of Moscow Institute of Physics and Technology headed by Professor Andrei Nikolaev and Professor Denis Lunev. Vilgelm Bitner, former employee at IBM and Parllels and currently HPC Hub CEO, initiated shaping the professional backbone of the new company and designing the HPCaaS (High Performance Computing as a Service) alpha release. He also invited Eugene Protasenko to join the project. By then, Eugene Protasenko already had some financial and entrepreneurial experience, so the former banker was put in charge of raising investments and strategy development.

The first attempt to attract investments failed, as investors found the project of a supercloud service too sophisticated and the need for HPC market virtualization questionable. A year later, when it became evident the market is advancing in the very direction the project enthusiasts expected, first investments were raised. The funds invested by Dmitry Mikhailov, CEO at EG Capital Partners, in spring of 2015, were used to develop the core technology which was presented in the market in 2016.

HPC Hub was devising its strategy at the accelerator of the Internet Initiatives Development Fund (IIDF) and thus by late 2016 few basic assumptions were drafted. It took further six months to test the assumptions and to find first customers. In summer of 2017 HPC Hub raised a further round of investments from a group of investors, which was partially used to get a grant from Skolkovo Innovation Center. The grant was spent on finalization of some key components of the technology.

Currently, one basic platform is available for HPC Hub clients at the servers of Rosatom’s subsidiary in Sarov with capacity of over 35 teraflops. In the near future one more platform will be available at the servers of the Russian Space Systems.

In fact, we do not ensure full load for the Rosatom capacities. We will expand out supply as demand grows. The prospect sites are Russian Space Systems, Skolkovo Institute of Science and Technology and possibly Moscow Polytechnic University”, Eugene Protasenko says.

At the same time, the company is pursuing implementation of private cloud solutions and hybrid cloud solutions for major clients with infrastructure of their own.

Genome and seismic activity computations

Computation hypercapacities are required in knowledge intensive sectors. HPCaaS may be useful in field modeling, seismic data procession, bioinformatics or innovative medicines development.

Among HPC Hub’s initial clients were Soyuzneftegazservice and bioinformatics laboratory of the Scientific Research Institute of Physical-Chemical Medicine. Knomics (microbiome research company developing disease prevention, diagnostics, and therapies with the help of human microbes) which is part of Atlas Biomed Group, started using HPC Hub’s supercomputers cloud platform to advance its genetic tests research project. Earlier, when pursuing its project of intestinal metagenome research, Knomics successfully tested computation capacities and software while analyzing thousands of Russian biomaterial samples or over 10 TByte of data. On the basis of the data collected, the researchers put forward a hypothesis of dependence of the psychological state of a person on activities of bacteria in intestinal tract (the researchers believe that bacteria can have impact on the brain and thus cause mental depression or even autism). To ensure proper results of the research, HPC Hub’s specialists transferred a metagenome pipeline (metagenome data processing software set developed by Knomics researchers) to the cloud and accelerated its performance eight-fold.

Genome research requires huge computation capacities. According to Kirill Kaem, Senior Vice-President for Innovations at Skolkovo Foundation, such computation capacities are scares in the global biomedical market.

There are not enough solutions which could help to gather and analyze in one single location the entire information on microbiota research, so that these vast data could be used at the output by everyone, ranging from patients to doctors and to businesses. The product of the team is at the interface of big data and biomedicine and it will be in great demand both in Russia and worldwide”, Kaem claims.

Eugene Protasenko confirms that foreign bioinformatics researchers have shown interest toward the supercloud service.

We plan to actively develop this area on the basis of our expertise and our understanding of the industry we got thanks to working with Knomics”, the HPC Hub cofounder says.

According to Eugene Protasenko, a lot of pilot projects are underway in oilfield servicing and fuel and energy industry in general. HPC Hub develops solutions jointly with software vendors and intends to make respective public announcement in a short while. Besides, in cooperation with Energozapas Ltd. and a number of Chinese companies, various tests in engineering computations are now underway.

Our company has just come to the market with complex solutions, but it is already clear we will have our own niche in several industries and will possibly become technology vendors and partner in other competing projects”, Protasenko claims.

A competitor for HPC Hub is T-Systems. It offers to B2B segment a service of performing resource-intensive computations on its hardware. Universities which own supercomputers are also ready to lease them. HPC Hub sees its future in providing a more flexible and more convenient service to the end customer.

Missed leadership?

HPC Hub founders think the main hardships at the initial stage of the project are due to the heavy inertness and conservatism of major customers who have the biggest money. For the same reasons, too much time is consumed when establishing new partnerships with the software vendors. Still, the market’s overall progress is toward the clouds and that allows the team to expect that in the very near future it will secure results it has been striving to get for the second year now.

Another serious challenge is the complexity of the market, so it took a very long time to persuade the first Russian investors the HPCaaS solutions have great outlooks. We had a chance to be the first in the global market, but that chance did not materialize”, Eugene Protasenko laments. “Now, when there are analogous products and competition in the market, the problem is in persuading the new investors that our team is not only able to compete, but to lead the global market too. In Russia, there are not enough funds to develop such hardcore technologies. That ultimately limits our possibilities at the offset. We are at the very offset indeed, and we have to choose the cheapest options and advance by tiny steps. So we advance at a slower pace compared to what it could be if we had investments at the scale of our competitors in Silicon Valley”.

HPC Hub is certain that if it had a comparable budget, it would be able to ensure much better results in technologies quality and solutions efficiency.

Even though Russia is not leading the rankings of the nations with superior computation capacities, its market of HPC capacities on demand has good prospects. In 2016 it was assessed at RUR 7.8 billion ($ 135 million).

There are several premises for that. First of all, computation clusters of major state corporations and universities use, at best, 30% of their capacities. The rest of the time their expensive equipment stands idle. Buying a standard 1,000 kernels cluster and its five years maintenance costs approximately RUR 100 million ($ 1.8 million). An equivalent time in a professional supercomputer cloud would cost half of that.

By Anna Oreshkina

Previous ArticleNext Article

1 Comment

  1. Gretings! I’ve been гeadіng your blog for a while now and finally got the courɑge to go ahead and give you a shout out fг᧐m Humbⅼe Texas!
    Just wanted to say keep up the good woгk!