Business challenge

The Technological Institute of Tijuana was preparing to do top research on many domains in the need for machine learning and optimization, including social sciences issues. Their x86 infrastructure did not have the performance and scalability they needed to perform the computations that such work requires.

Transformation

The Technological Institute of Tijuana worked with IBM to build an IT infrastructure for top research with IBM Cloud Private and Docker on IBM Power Systems S822LC for High Performance Computing, developing their own machine learning systems based on evolutionary computation.

Results

Over 100x faster

computational time vs. x86

Processing time reduced

from 4 hours to a minute

Scalability to over 1 million

data points and hundreds of thousands of features

Business challenge story

Building an IT Infrastructure for Top Research

The Technological Institute of Tijuana is an institution of higher education located in Tijuana, Baja California, Mexico. The institute was founded on September 17th, 1971 and was one of the first higher education institutions in the state. Between their two campuses, a total of 21 bachelors, 6 masters and 3 doctorates are offered, being a key source of professional and scientific talent in the region. Their function is to provide higher education services at undergraduate, masters and doctorate levels based on the skills needs in the region.

The institute has been doing research into real-world problems that can be tackled with machine learning techniques, such as human immigration patterns and other social science issues that are relevant to border cities between Mexico and the USA. They had a very large data set in which they were developing their own algorithms. These algorithms would be used to make predictions on human migration patterns and other issues affecting society. They began to determine whether or not their existing infrastructure had the capabilities needed to perform research on these issues at the highest level.

The institute had existing Dell servers with Intel Xeon processors for their research lab. The team came to the conclusion that these servers would not have the bandwidth and capabilities needed to perform the research they were looking to do with big data, particularly for performing complex optimization and learning processes. They needed accelerated servers with GPUs to handle high performance computing workloads. These servers would also have to scale up to meet the demands of their data sets.    

Dr. Leonardo Trujillo, a Research Professor at the Technological Institute of Tijuana in the Department of Electrical and Electronic Engineering, began researching IT infrastructure for high performance computing that would enable them to perform research at the highest level. They knew about the relationship between IBM and NVIDIA for enterprise AI and high performance computing, so they began to look at Power Systems to build out their research infrastructure. After speaking with IBM and researching Power Systems for high performance computing, the team was confident that this was the direction they needed to go. They began to build out their new research infrastructure with IBM Power Systems.      

Transformation story

Faster Computational Time and Increased Scalability

Working with IBM, the Technological Institute of Tijuana adopted IBM Cloud Private on Power Systems S822LC for High Performance Computing. Power Systems S822LC for High Performance Computing are accelerated servers with NVIDIA Tesla GPUs and NVLink technology for faster CPU to GPU communication. IBM Cloud Private on Power is an open Kubernetes platform that helps developers easily build and deploy containerized applications. With IBM Cloud Private on Power, Dr. Trujillo and his team can feel confident in deploying their high performance computing applications rapidly in an on-prem private cloud environment — securely and with the best availability.  

The institute also deployed Docker containers on their new IBM Power Systems infrastructure. Docker provides enterprise container software that is ideal for developers looking to get started with container-based applications. Using Docker on IBM Power Systems, Dr. Trujillo and his team can containerize their AI algorithms to do specific searches on large data sets. They can divide a search within a big data set into segments and launch a Docker container that will focus just on that specific search, which decreases the time spent looking for specific insights and allows them to do sophisticated state-of-the-art search patterns in a very simple and efficient manner.      

With IBM Cloud Private and Docker on Power Systems, the team has been able to reduce computational time for their research workloads from around 4 hours with their old Dell and Intel Xeon infrastructure down to just one or two minutes — a more than 100x increase in speed. Using NVIDIA NVLink technology on IBM Power Systems, data is transferred between the CPU and GPU faster than with a traditional architecture. IBM Power Systems have up to 5.6x more I/O bandwidth than x86 servers to reduce bottlenecks for high performance computing workloads. Dr. Trujillo and his team are also able to share RAM across CPUs and GPUs for even greater performance. The container-based approach also reduces their deployment time and learning curve for young researchers and students that join their lab.

Dr. Trujillo and his team are also able to scale up quicker and further than before with their old traditional infrastructure. Scalability is critical for the work they are performing, as it allows them to better characterize the dynamics of their search models in a more precise manner for research applications. With their old traditional infrastructure, they were only able to run workstations using 1,000 to 2,000 data search points. Now, they can scale to over a million data points with Power Systems. This scalability allows them to gain new insights into their data that were not previously possible. 

We work at the intersection of evolutionary computation and machine learning, a field called Genetic Programming, where it is possible to automatically generate impressive predictive tools using sophisticated search and optimization methods, but to do so a lot of computing power is required, and fortunately the IBM POWER8 HPC platform with NVIDIA GPUs provide more than enough.

Dr. Leonardo Trujillo, Research Professor, Technological Institute of Tijuana

Results story

Solving Social Sciences Issues

The Technological Institute of Tijuana now has the IT infrastructure they need to do top research on many real-world domains that require state-of-the-art machine learning and optimization techniques, including social sciences issues. In this area, for example, machine learning can allow Dr. Trujillo and their team to explore the immigration patterns of people that come to the border between Mexico and the USA, a particularly important problem for the city of Tijuana, since it is one of the largest and most important border cities in the world. IBM Power Systems have the performance they need for modern high performance computing and data-intensive workloads. With this technology, their goal is to generate computational models that can help classify and predict the immigration patterns of people that come to Tijuana and other border towns. They are also undertaking new research projects in related fields, such as evolving new AI algorithms from scratch, performing automatic code-repair and code improvement of software systems, understanding nanotechnology properties of new materials through machine learning, automatically generating computer vision algorithms, and developing powerful machine learning platforms for diverse domains. IBM Power Systems have the compute power they need as they take on other new research initiatives.   

Dr. Trujillo and his team were pleasantly surprised how fast the learning curve was to get started. In a short amount of time, they were able to get their new high performance computing platform up and running to get started on new research projects. They were quickly able to deploy their research algorithms onto the new infrastructure quickly to get started on data initiatives.

The institute is now working on their next step in implementing IBM Watson Machine Learning Accelerator. Watson Machine Learning Accelerator makes deep learning and machine learning more accessible by combining open source deep learning frameworks with efficient development tools and accelerated IBM Power Systems servers. With Watson Machine Learning Accelerator, Dr. Trujillo and his team will have a complete environment for data science, enabling them to perform even more complex and impactful research. They will also have the scalability they need to handle deep learning on increasingly large data sets.     

Furthermore, Dr. Leonardo Trujillo and his team are also looking at further expanding their partnerships and relationships with other research institutions and universities. With IBM Cloud Private and Power Systems at the center of their research infrastructure, collaboration with another institution will be even more effective than going at it alone. Their new infrastructure also helps solidify their status at a top research institution in Mexico and North America. The Technological Institute of Tijuana is looking forward to solving important social sciences issues with IBM Power Systems.   

With the help of IBM and NVIDIA GPUs, we can tackle large scale problems in many domains, exploring the space of possible solutions in a highly efficient manner and allowing us to build tools and gather insights that were previously out of our reach.

Dr. Leonardo Trujillo, Research Professor, Technological Institute of Tijuana

business logo

About Mexican Technical Institute/Technological Institute of Tijuana

The Technological Institute of Tijuana is an institution of higher education located in Tijuana, Baja California, Mexico. The institute was founded on September 17th, 1971 and was one of the first higher education institutions in the state. Between their two campuses, a total of 21 bachelors, 6 masters and 3 doctorates are offered, being a key source of professional and scientific talent in the region. Their function is to provide higher education services at undergraduate, masters and doctorate levels based on the skills needs in the region.

Solution components

Take the next step

To learn more about IBM Power Systems S822LC for High Performance Computing servers, please contact your IBM representative or IBM Business Partner.

To learn more about IBM Cloud Private on Power, please contact your IBM representative or IBM Business Partner, or https://www.ibm.com/us-en/marketplace/cloud-private-on-power

To learn more about IBM Watson Machine Learning Accelerator, please contact your IBM representative or IBM Business Partner, or https://www.ibm.com/us-en/marketplace/deep-learning-platform