Sowmya Venkataramani is a Program Director in the University Research and Collaboration group within Intel Labs, where she manages Intel's academic research investments in the areas of Artificial Intelligence and Neuromorphic Computing.
Highlights:
- In 2020, Intel Labs and its university partners established the Telemetry Data Science Center of Excellence (Telemetry CoE).
- This collaborative effort utilizes data points from millions of devices using longitudinal data over multiple years and has produced valuable insights and research efforts.
- CoE topics include studying power and cycles consumed by browser and web-based applications, building algorithms to discover PCs usage patterns, impacting user experience via both soft and hard errors, and determining the Scope-3 carbon footprint impact as required by Intel’s RISE 2030 goals.
- Intel Labs aims to expand the CoE by adding new academic and industry partners.
As the research arm of Intel, Intel Labs highly values university research partnerships. As such, Intel Labs has funded many research centers throughout the years. In 2020, Intel Labs funded the Telemetry Data Science Center of Excellence (Telemetry CoE). Built upon a small initial framework, the Telemetry CoE aims to foster collaboration between Intel subject matter experts, university students, and professors and build upon Intel’s technical pipeline of principal engineers to recruit future Intel employees via internships. Through these partnerships, the team applies data science techniques to create actionable insights and publish analytics, insights, and results for Intel and the industry.
“Faculty members and Intel technical leaders have collaborated to address tough Machine Learning challenges related to time-series analysis. Intel is helping us train a compelling talent pool of data scientists. We have built a strong bond together that will serve as an exemplar of effective research collaborations for the Institute.”
- Rajesh K. Gupta, PhD, Founding Director, Halıcıoğlu Data Science Institute, Distinguished Professor, Computer Science & Engineering, University of California, San Diego
“Solving hard problems requires a myriad of skills brought together by highly collaborative people. Our University Data Science CoE enables meaningful problems to be addressed while training the next generation of talent that can help Intel succeed in the marketplace.”
-Bijan Arbab, Director of Telemetry Strategy at Intel
Telemetry Data
Telemetry is the automatic measurement and transmission of data from remote sources. Through a number of sensors, telemetry gives technology professionals the ability to observe components and monitor applications in a deeper way, with metrics that track performance, utilization, energy consumption, and more. By making use of telemetry data, companies can improve key aspects of their organization, including reliability, security, performance, and power consumption. Telemetry can also generate insights to help Intel and other vendors/manufacturers evolve capacity requirements, improve user experiences, and help prevent system errors.
Intel is very focused on delivering positive user experiences and reliable performance. Telemetry data plays a large role in making improvements over each iteration of Intel products, as it provides data center operators with a deeper understanding of how the products perform in the real world. Analyzing the data can alert Intel and its partners to errors and aid in finding the root cause. For example, does a system really offer 15 hours of battery life? Does it really wake in less than a second, as promised by instant wake? Does it overheat? Do certain programs impact CPU and memory more than they should? Overall, does the system perform the same outside of the lab?
Data Collection and Analytics
Many users opt-in to send their data for analysis. In fact, Intel systems have an opt-in rate of 30-40%, meaning that over 6 million consumers send data daily, and Intel collects telemetry data from 39 million systems, a number that grows every day. While this data is incredibly helpful, it can be overwhelming on such a large scale. One error multiplied by 39 million users over time generates a petabyte of data - so much data that it had to be offloaded from Intel’s IT system to Amazon Web Services.
Traditional methodology of designing, building, pricing, and selling devices to standard industry benchmarks is not effective since real user experiences are impossible to capture via standard benchmarks. Real user experiences are dynamic and ever-changing, and as such, it is difficult for industry benchmarks to keep up with the dynamic and ever-changing nature of actual device usage en masse.
The Data Collection & Analytics (DCA) team at Intel analyzes real-world telemetry data to determine system health across CPU, RAM, storage, Wi-Fi, web usage, power consumption, load times, programs, event logs, and thermals. The DCA team is comprised of talented engineers and data scientists, but with such an extraordinary amount of data to process, they needed help. In addition, no single telemetry solution provided by a single eco-system partner has a high probability of mass adoption. However, a consortium-based telemetry solution led by a neutral party (such as a university) has a high probability of success and adoption by the rest of the industry ecosystem players.
With this in mind, the team reached out to the University of California San Diego (UCSD) to partner with professors and students who are studying the latest methods of data science. To ensure data privacy and security, the DCA team implemented a strong legal framework for sharing device telemetry data to protect Intel and our customers. The program has been extended to the University of Washington, another highly-ranked data science university.
Program Results
As of 2022, the Telemetry CoE has produced 13 white papers and 18 studies. Additionally, the program engaged 15 engineers, principal engineers (PEs), senior principal engineers (Sr. PEs), and fellows, and Intel took on 12 interns* and hired six other students as full-time employees.
*Note: Internships and hiring were conducted prior to the current fiscal environment.
Some of the Telemetry CoE focuses include:
- Happiness Project - identifying sources and root causes of everything from slow system responses to bluescreen-of-death.
- Web-Based Computing - studying power and cycles consumed by browser and web-based applications, which are now responsible for more than 60% of PC CPU cycles worldwide.
- Impact of PC Usage on Users Health - building algorithms to discover PC usage patterns, which also allows us to discover engagement levels with various Intel versus competitor’s SW tools (Compilers, Debuggers, Optimizers, etc.).
- Causal Analysis of Silicon Failures - impacting user experience via both soft and hard errors.
- COVID Time Series Analysis - enabling us to better understand and predict PC usage patterns.
- Carbon Impact Studies - determining the Scope-3 carbon footprint impact as required by Intel’s RISE 2030 goals.
One of Intel’s pillars of Fearless Innovation is to “continuously improve, enabling our teams to be more curious, bold, and innovative.” The Telemetry CoE engages the brightest minds at the universities and connects students with Intel veterans to solve real-world problems by securely using Intel’s telemetry data. Focused on driving innovation and results, this program has fostered world-class collaborations, produced valuable research publications, and uncovered great data insights for the industry at large.
The team looks forward to growing the Telemetry CoE’s impact through collaborations with other universities and industry partners. It is now time to scale and invite other industry partners to join in our collaborative telemetry-based data-sharing consortium with the goal of developing actionable insights leading to better end-user experiences that allows all eco-system partners to effectively compete with single vertical-oriented device providers. Come join Intel and the University Telemetry Consortium Center of Excellence!
For more information on how to join the CoE, contact Bijan Arbab at bijan.arbab@intel.com.
Sowmya Venkataramani has a Ph.D in physics from Brown University. Funded by DARPA Bio-Info-Micro Program, her dissertation research focused on designing and developing InGaN/AlInGn light emitting diode arrays (LED’s) for dynamic imaging of neural circuitry. Sowmya successfully cultured and patterned two dimensional networks of hippocampal neurons using novel micro-contact printing techniques and used fluorescence-based imaging techniques to study signal transmission in these artificially patterned networks. Sowmya was a recipient of the Burroughs Welcome Fellowship.
As a post-doc at UC San Diego, Sowmya developed two-photon imaging methods to study synaptic plasticity in cortical neurons. Working at the Casey Eye Institute, at OHSU, Sowmya studied synaptic mechanisms involved in visual processing, in particular studying direction and orientation selectivity in retinal ganglion cells.
Building on her diverse background in Physics and Neurophysiology, Sowmya has worked in the field of patents and tech transfer, where she has served as a patent specialist and scientific advisor, advising on inventions in the fields of medical devices, optical systems, driver assistance systems, and engine systems for Fortune 500 companies and Universities.
At Intel, Sowmya first started in the Global Supply Chain organization, where she was a Technical Program Manager managing key metrology programs and driving collaborations with suppliers and developing solutions to meet Intel’s roadmaps. Currently, she is a Program Director in the University Research and Collaboration group in Intel Labs where she manages Intel's academic research investments in the areas of Artificial Intelligence and Neuromorphic Computing. She is also Program Director of Intel’s Semiconductor Education and Research Program for Ohio.