Kaylen Thorpe

#Linux/Unix #Windows #networking #Amazon AWS #monitoring #development #databases #storage #virtualization #scripting #security

ADP (2019 - 2024)

• Created performance testing automation and orchestration framework in Python that runs in Kubernetes
  
  • Supports multiple performance testing tools like Locust, JMeter, and K6
  • Orchestrates worker pods and nodes, collects, analyzes, reports, and persists output
  • Integrates into CICD pipeline to automate performance testing and identify regressions
  • Created SDK to enable development teams to create and automate their own tests
  
  
• Spent extensive time on performance testing and tuning of AWS cloud environment
  
  • Tested numerous types of AWS and Kubernetes resources, software services, APIs, and more
  • Benchmark various instance types/sizes for services like EC2, RDS, EKS, and ElasticSearch
  • Determine the most performance and cost-efficient instance types to use for EC2, RDS, EKS, and ElasticSearch
  • Reverse-engineered and automated the performance testing of our payroll calculation system. 'Nuff said.
  
  
• Cost optimization of Kubernetes clusters
  
  • Create datadog dashboard to calculate/report on service and cluster utilization
  • Adopted self-service architecture; document how to use this tool to optimize a service, demoed to entire department
  • Worked with service owners to analyze the performance profile of their services and make informed recommendations
  • Reduced the number of nodes running in the cluster and increased overall utilization, thereby reducing expense.
  
  
• Created/implemented an application performance monitoring library in NodeJs across entire platform using Datadog APM, with distributed tracing capabilities, to allow developers to correlate log events with detailed performance data, and troubleshoot and optimize their code.

PEX Card (2015)

• High-performance Microsoft SQL Server database cluster
  
  
• Designed and installed Dell Compellent Storage Center iSCSI SAN with 10G dedicated network and RAID 1
  
  
• Built a highly resiliant, distributed Active Directory environment with physically diverse servers that utilized Windows Server Update Services (WSUS), Windows Deployent Services (WDS), Read-Only Directory Controllers, Group Policy and nxlog among other tools for centralized administration.

PEX Card (2015)
Worked with business analysts to generate and import financial datasets, create live visualizations, and a graphical dashboard to display throughout the office.

CUNY Macaulay Honors College (2015)
Student activities fund online application, administrative review, and follow-up documentation and tasks.

City of New York City Clerk's Office Marriage Bureau (2015)
Implemented the OpenKM document versioning system to replace unreliable Subversion. Worked with the New York City DoIT (Department of IT) to deploy securely onto the New York City intranet.

City of New York City Clerk's Office Lobbying Bureau (2014)
Database of current and former (imported, manual entry) individuals and applicants who are certified to notarize documents within the 5 boroughs. Generates bulk and individual letters for legislative approval, notifications and renewals, and prints official cards on custom designed paper.

CUNY Queensboro - LaGuardia & Wagner Archives (2012)
"Fact Of The Day" database & iOS app

NYC Community Air Survey - Queens College (2011)
Normalize and import raw data of various formats into PostgreSQL database, quality control of data, reporting, future schedules.

CUNY IT Specialists Internship Program (2010-2013)
Employment management application written from the ground-up using Object-Oriented PHP5, managed the application and hiring process, improved work assignments closer to home by plotting employers and employees on a map, tracked and reported hours and electronic timesheets, and generated myriad reports.

Daily Orange Advertising Database (2006-2009)
Customer-relations management application written in PHP and MySQL.

Bitcoin market monitoring, analysis and alerting tools (2013)
Wrote a fully-functioning, RFC-compliant, socket-based IRC server (which accepted persistent client connections) and administrative bot using only mIRC script (2004)
Blacknova Traders (open-source turn-based online game in PHP/MySQL) (2003)
IRC weather reporting bot (2002)
Multi-tempo, multi-octave music composer program written in BASIC, run on DOS 6, and output through the PC speaker (1997)

"Big Data" - John Jay College Building Automation System

In 2012, the CUNY Institute for Software Design and Development (CISDD) partnered with the CUNY Building Performance Lab (BPL) and John Jay College (CUNY) to begin a study of the efficiency of the HVAC (heating, ventilation, and cooling) system of their newly-unveiled 12-story high-rise tower. Each group had a goal: CISDD's ultimate goal was to create a machine learning system that could predict the optimal settings for the system. BPL sought to develop, test, and document new methods of reducing utility usage. And John Jay sought to reduce utility usage, improve efficiency and responsiveness, and justify additional expenses. This presented several challenges: collecting the data, importing it into our system, storage to keep it on, a database powerful enough to query efficiently, and the software to analyze and report it with.

Opened in the fall of 2011, the John Jay College Extension is a 12-story glass tower with 4 basement levels. It's "building automation system" (BAS) was manufactured and installed by Siemens, which controls all segments of the heating/cooling system, including the cooling towers and chillers on the roof, the outside air intake in the "penthouse", the air handlers and damper controllers on each floor, and the steam heating system. The system is controlled by creating a policy on an administration server/console, and pushing that policy to the "field control units" over a TCP/IP management network. The field control unit implements the policy, and controls the actual opening and closing of the valves/circuits to meet the temperature setpoint for that zone in response to its temperature sensor data. The server/console used by the building engineer only performs two functions: manages and pushes temperature setpoints and control policies, and displays a live overview of current system sensor values; it is a very basic interface and application. Since the intelligence of the building automation system lies at the edge (field control units), these devices only have the memory space to collect a very limited number of datapoints, and they must be polled by the server/workstation by a report, that can only be run manually.

We sniffed the BAS management network to try and intercept sensor data packets, but were unable to decipher them. The Building Performance Lab has been working with Siemens to try and collect those packets independently, and the idea is to develop an appliance that sits on the network and live-polls all (requested) sensors. In the meantime, the only means to collect consistent data is to have an intern on-site to manually transfer the data to CISDD servers. A script that I developed then sanitizes, normalizes, catalogs, generalizes, and imports that data into a MySQL database. Towards the end of my tenure, we were exploring the MongoDB NoSQL database to expand our data model to new buildings, devices, and sensors at other campuses, like our new partnership with the City College of New York (CUNY).

The most fun part of the project for me was developing the software that we would use to collect and analyze the data. I built an entire catalog and visualization system that would import report files and record datapoints as a model of 'report'->'dataset'->'sensor'->'datapoint'. Using the d3, Highcharts, Chart.js, and jQuery libraries, I built a responsive AJAX application that organized buildings, floors, sensors, reports, and datasets in an easy-to-use interface. It can visualize data in many different types of formats, and generate visualizations and reports based on user-defined parameters. Some reports were tailored to output in a unique format for use with a different application used by the BPL.
Some of these novel visualizations included:

• Using a scatter plot to correlate sensors with a similar range of values. The minimum value is plotted on the X scale and the maximum value is plotted on the Y scale. It is most useful to plot this on two separate graphs: one using a logarithmic scale to identify sensors with high values (like 100,000 KWh), and one on a normal scale with maximum values of, say, 10000. For example, a sensor whose values range between -40 and 130, or especially between 0 and 110 is probably a temperature sensor. A sensor that has values between 0 and 100 is probably a percentage sensor, which probably indicates the amount that a valve/damper is open or closed or the percentage of maximum speed that a motor is operating at. A sensor that is in the +100,000 range is probably cumulative electrical usage of major system components, like the cooling towers on the roof.
• Use a 12-month calendar to visualize data health/completeness by using darker/lighter colors to indicate how many datapoints are available for each day.
• Use a rectangular heatmap to represent 0-24 hours / 365 days, to visualize the heating and cooling trends that take place from morning to night and from season to season, for every temperature sensor. This also visualizes the different heating/cooling patterns that take place on regular class days or on weekend/vacation days when there are fewer people in the building.

The BAS system has a lagged responsive time - it can take +15 minutes for the temperature of a room to change once a damper is opened. We want to make the room comfortable for occupants, which may be a setpoint temperature of 72 degrees, but it might be 80 outside and presently 78 in the room. Since we do not want to waste energy, we would begin cooling the room to the 72 degree setpoint, 20 minutes before the scheduled occupancy time. A large conference room, lecture hall, or atrium requires extra HVAC resources to handle the greater volume of air and larger number of people, and takes longer for air to circulate.

Air handler units (AHUs) are large blowers (typically 3 or 4 per floor) that contain several filters, and distribute fresh air coming from the outside air intake on the roof to a section of ducts. One novel way that we found to apply big data analysis was to identify dirty filters by comparing the air pressure on each side of those filters - a large discrepancy probably indicates a dirty filter.

An eventual goal of this project was to use predictive analysis and machine learning to try and use many years of building and weather data to estimate the utility usage and optimal system settings for use on hot/cold days, and in relation to solar exposure (the sides of the building on which the sun shines heat up more than the sides that do not, and at different times of day). I worked closely with our PhD researchers to define the best methods and datapoints to compute and best problems to try and tackle.

I created two APIs for our researchers: one that output JSON data describing the objects in the system (building, report, dataset, sensor, datapoint), and another that allowed them to run queries and custom reports, and returned values in CVS or JSON format. They used this to construct an elaborate R application that would output various scenarios based on historical data and recommend optimized setpoint values to reduce utility usage and speed up system responsiveness.

---

October 2013
Montreal, Quebec

Operational Effectiveness in Use of BAS (Building Automation Systems) [PDF]

Int'l Conference for Enhanced Building Operations
Bobker M, Brown T, Thorpe K, Joseph M, Ascazubi M, Essman J

NYC Taxi & Limousine Commission (TLC)

Trip Efficiency Study Use taxi trip data (one record per cab ride) to determine if cab drivers are taking the most efficient route, or if they are overcharging passengers.

Trip Search Engine Create a web interface and application controller capable of search of +1 billion records

Taxicab messaging system Created a web application through which the TLC can send text messages to taxi cab consoles based on their geographic (GPS) position to alert on traffic, accidents, and other important information.

Fire Department of New York (FDNY)

Civilian casualty report Electronic workflow efficacy to replace the 10+ page paper report that must be completed and approved by 5 ranks of officers, for each civilian casualty in an FDNY-involved incident.

NYC Metropolitan Transit Authority (MTA)

Express bus service How long can a traffic light remain yellow for a bus to pass through, without impact to the surrounding traffic grid? Analyzed the bus trip logs of the M1 Manhattan express bus over several months to determine constraints, rates of speed, and model traffic light timing under different scenarios.

CUNY - Queensboro Community College

Solar reflectance Wrote a proposal on tools and methods to collect solar reflectance data from the rooftop of Queensboro Community College, then determine the ideal placement of solar panels. The tools and methods that were developed were demonstrated to and shared with professors and officials at other universities to apply and improve upon.

CUNY - John Jay College

Building HVAC automation system Extensive "big data" research project that covered everything from the firmware on the data collection sensors in the ceiling, to the normalization and import of that data into various formats/databases, the development of a robust visualization and analysis application using the 'd3' and 'Highcharts' libraries among others. Generated a wide variety of visualizations and reports, managed datasets, tracked weather data, and much more. Utilized SQL and NOSQL solutions including mongodb. Worked with doctoral researchers within my department as well as at the CUNY Institute for Urban Systems to start implementation of machine learning and predictive analysis. This is a totally awesome project!