Teralab: AI Infrastructure for Research

Teralab: AI Infrastructure for Research

Nexedi: largest OSS publisher in Europe

Nexedi Office Map

Nexedi is the largest publisher of Free Software in Europe with a portfolio of more 10 enterprise software and customers such as Airbus, SANEF, Mitsubishi, etc. Nexedi is currently operating from offices in Lille, Paris, Munich, Shanghai and Tokyo. For more information, please refer to www.nexedi.com

Stack: 100% Free Software (no US restrictions)

Nexedi software stack covers all the needs of the enterprise: infrastructure, applications and productivity. It is licensed as Free Software with no US content restrictions and can thus be deployed anywhere in the world independently of political instability. Please refer to the Nexedi Free Software stack for more information on individual and complimentary solutions.

Problem (2012)

  • Large corporations have data
  • but have no data scientists
  • and no infrastructure
  • and they do not trust Amazon

The story of Teralab is not a story of Nexedi. Teralab has been created by business developers and engineers of Institut Mines Telecom since 2012: Pierre Pleven, Philippe Letellier, Anne-Sophie Taillandier, Olivier Dehoux, Gilles Polart-Donat and many others which I am not aware of. I appologize for not mentioning their name here.

In 2012, French large corporations and government agencies were looking for a platform that could be used to start implementing Big Data and AI algorithms. However, they did not trust private sector to keep of copy of their most secret data. Therefore, they did not want to rely on companies such as Amazon or OVH.

Those French large corporations and government agencies did not either have access to sufficient internal IT resources. Most datacenters of large corporations were optimised for business applications, not for big data. 

Last, they did not even have data scientists among their staff.

They were thus looking for a solution to start big data and AI experiments, in a secure way, without infrastructure and without data scientists.

Solution (2014)

  • French ministry of economy hosts infrastructure
  • Only Free Software without backdoors
  • Multi-level security
  • Support from academic data science teams

In 2014, Teralab started offering services to dozens of large corporations and agencies. Each project consists of a cluster of virtual machines sharing sensitive data. Clusters are mutually isolated thanks to 3 levels of networking filtering, handled by two independent human teams. It is thus difficult to break Teralab's isolation thanks to redundant design, both from a social and technical way.

Teralab is hosted in one of the engineering schools under supervisions of the ministry of economy. 

All its software is Free Software: this removes the risk of hidden backdoors found in proprietary solutions. Even routing is implemented using Free Software so that there is no dependency to any form of proprietary hardware from companies such as CISCO, Juniper or Huawei. The whole infrastructure's source code can be audited..

Teralab supported its users thanks to the collaboration of teams of data scientists part of Insititut Mines Telecom (IMT) group of engineering schools.

Result (2016)

  • Large corporations reach success with AI
  • Data scientists access data and create AI
  • Government owns a new profitable company

In 2016, Teralab reached complete success. Dozens of French corporations and government agencies could implement innovative big data, machine learning and AI processing algorithms on their data. This success was possible thanks to the combination of trusted infrastructure, business developers and researcher teams to support Teralab users.

A European Success (2016)

Teralab was thus awarded in 2016 the "Silver i-Space” label form the European Big Data Value Association. The first success story of Teralab was published: a big data platform with a European vision.

A Public Policy Success (2018)

In 2018, Teralab was presented at Vivatech exhibition to French Minister of Economy: Bruno Lemaire. Its operation has reached break-even. It is considered as a success of public policy.

Key Technical Success Factors

  • SlapOS (not OpenStack)
  • Low cost hardware (not IBM)
  • Routing (not switching)
  • DAS (not SAN)
  • Very high Free Software Expertise (not Microsoft)
  • Python (not only Java)

Teralab is not primarily a technical success. But without sound technology, it would not have succeeded either. Other projects initiated by French Government have not always made such sound technology choice and did fail. It is thus useful to understand why Teralab could succeed from a technical point of view.

Teralab is based on SlapOS, Nexedi's Free Software for edge computing and cloud orchestration which provides stability of performance. It is not based on OpenStack, which stability issues have become notorious over years

It relies primarily on low cost hardware from companies such as Supermicro or from the Open Computer Platform. It does not rely on expensive servers from premium brands such as IBM. It is thus competitive from a price point of view.

It used routing only implemented using babel protocol. It does not rely on switching. This simplifies greatly the implementation of multi-site clusters or the integration of remote infrastructures.

It uses direct attached storage for best I/O performance and lowest latency. It does not use storage area network technology.

It is managed by engineers with long Linux and Free Software experience. It does not rely on any Microsoft technology.

It uses a lot python, and not only Java based technologies.

Nexedi AI Infrastructure

  • Lowest cost
  • Looking forward to extend it in China

The success of Teralab has convinced Nexedi to commercialise a similar setup for small and medium businesses based on a monthly fee. Initial servers have been installed in June 2018 in a datacenter in France. Their cost will be 30% lower than lowest European market price and possibly more than 5 times cheaper than the cheapest similar offer in China. We hope that this offer will help small and medium business to adopt big data and AI technologies. We expect to build strong partnerships in China to extend our offer since we believe that AI can only be widely adopted with competitive infrastructure.

Thank You

Image Nexedi Office
  • Nexedi SA
  • 147 Rue du Ballon
  • 59110 La Madeleine
  • France