Developed an in-memory file system using the FUSE library . The FUSE library captures all the system calls, and returns them to the user space. I wrote implementations for the basic system calls like read, write, open, close, etc.

Developed a non preemptive threading library using context based system calls (setcontext, makecontext, swapcontext, getcontext) which supported create, yield, join, and exit thread routines. Supported semaphore routines - init, signal, wait, destroy.

Developed a micro command line shell which accepted all linux commands. Implemented custom built-in functions for a few commands - cd, echo, logout, nice, pwd, setenv, unsetenv, where. Supported I/O redirection and inter-process communication using the pipe command and duplication of file descriptors.

Developed a network of computers where each player would play the Hot Potato game. The modified version of the game consisted of a packet being passed around the ring, with a specified number of hops, where each player would append its player id to the packet string. As soon as the number of hops reduces to 0, the player with the packet will send the packet back to the master.

Implemented reliable file transfer between two machines using UDP sockets. Achieved reliability using the Go-Back-N ARQ protocol. Used threads for retransmission timer and for packet resend functionality.

Implemented a simple peer to peer system for downloading RFCs. Maintained a centralized index on a known server which contained information about the location of RFCs on the network.

A database application which handles all the housing and parking options for students and registered guests. Developed the modules for interacting with the database using JDBC and wrote stored procedures to dynamically generate invoices and find a student a suitable roommate.

Improved and optimized SimpleDB buffer pool search by using HashMap instead of a linear search. Added support for new data types by adding get and set methods - Short, Boolean, Byte, Date.

End to end software delivery pipeline demonstrating the three prime stages of a software development project - Build, Test and Analysis, and Delivery. Used Git and Jenkins for the build process by configuring git hooks. Used Mocha and Istanbul for unit tests and code coverage measurement. For deployment, we developed a tool to automatically provision and configure a machine on AWS. Configured the machine using Ansible. Used Express.js to build a sample web application and demonstrated a canary release using http-proxy and Redis feature flags. As a special milestone, we implemented a doctor monkey and a maintenance monkey by developing tools to periodically check the system for irregularities and fix them in place.

Developed a tool to interact with the REST APIs of AWS and DigitalOcean and automatically provision a virtual machine on these platforms. Configured the machines using Ansible.

Developed a tool to analyze a source code's Abstract Syntax Tree (AST), and generate unit test cases which achieve 100% code coverage. Improved statement coverage from 71% to 100% and branch coverage from 40% to 100%.

Demonstrated three advanced docker features.

• File I/O - Exposed a port on the docker container to listen to TCP requests. Linked ambassador container to this container. Accessed a file on the original container through the ambassador container using curl.
• Ambassador pattern - Used docker-compose to spin up containers which were running a redis server and client. Linked two ambassador containers, one each to the client and server. The redis client and server communicated only through the ambassador, and never directly.
• Docker deploy - Demonstrated blue green deployment using docker containers. Configured a post-commit git hook which contained the commands to build a docker image and push it to the local docker repository.

Created an order-k (k >= 3) B-Tree which supported insert and search operations. Made the B-Tree persistent by storing it on disk on program termination. The tree would be loaded again when the program started.

Implemented three main types of disk based merge sort.

• Basic merge sort - Divided the input file into sorted runs based on memory size and merged all the sorted runs together to produce the final sorted output.
• Multi step merge sort - Divided the input file into sorted runs based on memory size. Merged the runs into sorted super runs. Merged the sorted super runs to produce the sorted final output. Achieved significant improvement in performance over basic merge sort for large inputs.
• Replacement selection merge sort - Used a min heap to create sorted runs. The runs were created by comparing the smallest element in memory with the first element on disk. The smaller element was stored in the secondary heap, and used for the next run. Run was created as soon as primary heap was empty. Merged sorted runs to produce the final output. Achieved improvement in performance over basic merge sort.

Simulated in-memory database indexes and availability lists. Supported three memory management techniques - first fit, best fit, and worst fit. Made the indexes persistent by storing them on disk on program termination.