Dr. Ravish Khosla
ravishkhosla (at) alumni (dot) purdue (dot) edu
"You see things and you say 'Why?'. But I dream things that never were and I say 'Why not?'" -- George Bernard Shaw
I am an experienced product manager/owner specializing in building complex and innovative software products. I have a doctorate degree in ECE/CS from Purdue University plus 6 years product management and leadership experience delivering customer facing products in Fortune 500 company. My project management experience ranges from small features to very large, 100 man year projects that have consistently exceeded customer expectations. What distinguishes me is my meticulous attention to detail and product quality, which helped launch multiple releases of 5 products on time while reducing bug backlog by 97%.
I am interested in product/project management and technical management/lead industry positions in software. I am also interested in any research that leads to ubiquitous communications. I like the dynamic and the business driven focus of a technical field. The ability to produce profit out of a technical invention or a financial trade is a great motivator for me. Studying financial markets being my hobby, I am also open to quantitative and trading positions in the financial industry.
My most recent job experience was in Ericsson Silicon Valley as a Product Owner/Manager/Technical Lead for the IP Routers portfolio including both physical and virtual products and management systems. I launched the next generation customer facing interfaces of the routers integrated with element and network management systems. This implemented my multi-year visionary strategy for evolving the system design of IP OAM with roadmap milestones. I had also defined minimum marketable feature (MMF) prioritized feature sets for customer demos and product releases on the roadmap. My work also included guiding customer facing teams in securing new contracts and expanding business along with solving customer issues.
I have experience leading a team of 200 system engineers and cross-functional teams spread over 5 countries who worked on multiple releases of 5 products. I have driven hundreds of small and large features through various stages, from planning to development to delivery. I championed an improved way of working based on Agile methodologies which increased productivity by 50%. I have also chaired the change control board (CCB) for IP Routers which manages change management for the entire portfolio. I have also been involved in staffing, recruitment and mentoring.
I believe in this quote by Artemus Ward, "It ain’t so much the things we don’t know that get us into trouble. It’s the things we know that just ain’t so." I enjoy picking up new skills and using them along with the immense information available from various sources to solve anything. At the same time, I believe in the quality of the skills learnt, not the quantity, and a scientific attitude in problem solving to avoid false conclusions.
My research interests span both wired and wireless networks. Throughout my career, I have been involved in communications and networking. Truly ubiquitous communications i.e. the ability to communicate and interact with anyone without the constraints of geography or mobility is at the core of my interests. Seamless communication between devices has been sought after for long but has yet to be realized at a commercial level. I am interested in any research that advances us to the goal of ubiquitous communication. I am also interested in machine learning and data mining as research techniques. I learnt about the field in a graduate course at Purdue University and have subsequently applied it in my research to predict IP prefix availability. I feel that these techniques are very relevant in making smart, scientific decisions given the immense amount of data available in many contexts.
My other hobbies include studying the financial markets. I have been studying stock markets since I was 15 years old. I have witnessed the 2008 financial crisis and the subsequent stock market crash along with the flash crashes that happen frequently.
MAC layer Design of a Peer-Peer Network: This project was undertaken during my internship at Qualcomm Corporate R&D and it involved Peer to Peer Networks. The internship was an amazing experience in System Design where I had to understand the current design of a system and then design mechanisms that fit into the existing design. The goal was to create a network where various user devices, each potentially operating on a different technology like 802.11 or Bluetooth, can discover each other and communicate. This has potential applications in improving the quality of life of the user, who does not have to deal with transferring his/her data to multiple devices. Other applications include a user's device remembering a user's preferences when he/she visits some public place e.g. a restaurant. I worked on a design that bridges disjoint subnetworks, which are created by the use of various technologies. This involved designing communication protocols and message formats, which had to fit in the existing layered design of the standard. My design was integrated with the current design and I received excellent feedback from executives. A patent on this design was granted in 2014.
Analyzing Prefix Reachability in the Internet: The Internet is composed of a complex set of organizations (Autonomous Systems or ASes in short) each setting its own policies for traffic flow through it. These policies can interact in unexpected ways causing unreachability of an IP address from portions of the Internet. Also, Border Gateway Protocol (BGP) routing instability, caused due to link failures or path oscillations, can lower the availability of IP prefixes (aggregated IP addresses). Connectivity in the Internet does not imply reachability and the reasons for reachability of IP addresses under BGP are not well understood. I studied how IP prefixes in the Internet become reachable and unreachable from various points in the Internet and explored various data-processing filters that can be applied to publicly available routing data to compute the average reachability of a prefix. The results show that the design of these filters and their parameters need to be tuned according to the needs of the study. Details of this can be found in my ICCCN 2009 paper, which can be found here.
Predicting Prefix Availability in the Internet: My published work in Infocom 2010 Mini-Conference (found here) focuses on predicting prefix availability in the Internet by building data-mining based prediction models constructed using the attributes of some Internet prefixes and applying it to predict the availability of completely unrelated prefixes. A clear tradeoff between the complexity of the prediction model and the prediction accuracy desired has been demonstrated. I investigated various easily-computable features of an Internet prefix which correlate with its availability. I also studied three different prediction models in detail and investigated the effect of the learning duration versus the prediction duration in the extended version of this work, which has been published in Elsevier Computer Networks journal and can be found here.
BGP Molecules: Understanding and Predicting IP Prefix failures: My published work in Infocom 2011 Mini-Conference (found here) is on studying correlation between failures or unreachability incidences across prefixes in the Internet. I investigated various features of Internet prefixes that make them similar w.r.t. their propensity to fail. I coined the term BGP molecule, which is the set of prefixes that are likely to fail along with a prefix of interest. This is a generalization of BGP atoms, which were proposed by researchers in 2002. The BGP molecules are then used to predict prefix failures using prediction models which consider AS paths and geographic location as prefix similarity features. My prediction technique predicted failures of 99.3% of prefixes of a random prefix sample with 91% accuracy. I then studied an application of BGP molecules aimed at enhancing the control-plane availability of servers returned by a Content Distribution Network (CDN) like Akamai. My study demonstrated that a significant percentage of queries for Akamai-hosted content exhibit lower than five-nines availability over a period of a few weeks. I investigated how the content servers returned by Akamai are related to BGP molecules and proposed a technique to increase availability by guarding against simultaneous reachability failures of primary and backup content servers. I showed that this scheme not only results in a near-perfect 100% availability for all content queries, but also has the potential of reducing client-perceived latency to the servers returned.
Cloud Computing: Interaction among Clouds: I also worked on Cloud Computing during my PhD. The ever-increasing demand for data and the cost-effective nature of outsourcing storage and computation to the cloud have led to the formation of increasingly large data centers by several organizations. Another important Internet trend is the increasing amount of content being delivered through it, dominated by live audio, video streaming and social networking. Increased presence of content providers is leading to a flatter Internet with content being delivered directly to customers from the provider clouds. I have formulated a model of the future cloud-based Internet and developed an evolution path to the Internet of tomorrow. I have also investigated upcoming challenges in the future Internet by studying various interactions or lightning between the clouds. My position on the outsourcing of services to the cloud is that of caution - blindly moving all services to the cloud can backfire; one needs to carefully evaluate the pros and cons of the move. While a flatter cloud-based Internet simplifies a lot of the current Internet's problems, significant issues need to be addressed in the future Internet by the networking community. Details of this project can be found in Chapter 6 of my dissertation.
Content Retrieval using Cloud-based DNS: Recent trends in cloud computing have moved DNS services to the cloud with several providers like Google and OpenDNS providing external DNS services. External DNS systems are worse than native DNS provided by an ISP for resolving highly distributed websites like Akamai. This is because the content server returned to a client when resolving Akamai hosted content through external DNS systems is far away from the client as compared to the one returned by resolution through local DNS. I have demonstrated that the problem is rooted in the disparity between the number and location of servers of the two cloud providers, that providing content (Akamai) and the one providing DNS services. I have developed a hybrid client-cloud technique that a client can use in today's Internet to locate a closeby Akamai content server while using external DNS, thereby preserving the benefits of using these DNS systems like security. I have also proposed futuristic cloud-based solutions of this problem, that will result in better performance as seen by a cloud user. This work has been published in the Global Internet Symposium 2012 and can be found here .
Study of Google public DNS: I also studied external DNS systems using Google DNS as a case study and evaluated the effectiveness of Google DNS in redirecting clients to the nearest data center. The results show that typically one can find a Google data center with a better latency than the one the client is redirected to. This indicates that traffic and data center load are given more priority in redirecting users to data centers. I performed this evaluation with no prior knowledge of Google data center locations. The locations were inferred using a novel technique based on my measurements. This technique for locating data centers can be applied to other cloud providers which use IP anycast to redirect clients to the nearest data center. Details of this project can be found in Chapter 5 of my dissertation.
Fault Tolerant Data Dissemination Protocols in Sensor Networks: I was involved with Sensor Networks during my Masters at Purdue University, where I worked on data dissemination protocols in sensor networks. I designed a new data dissemination protocol, SPMS-Rec which reduced energy consumption, while reducing the delay and maintaining reliability. I studied this and other push-pull protocols in my publication in WCNC 2007, which can be found here. I simulated the new protocol and existing state of the art protocols on GTSNetS, which was a significant software project because of the undocumented nature of the simulator. I spent some time understanding the layered implementation of the simulator and then modifying it to enable a more realistic simulation of a sensor network. This was my first experience on a huge software project in Object Oriented Programming and honed my C++ skills. In a related collaborative project, we studied analytically the tradeoff between multi-hop unicast and single hop broadcast of large data packets in sensor networks and showed that the decision to choose one over the other will depend on the fraction of nodes interested in obtaining the data. Details can be found in my publication in VTC 2007, which can be found here. I also designed a data dissemination protocol for sensor networks with malicious or faulty nodes, which uses the joint metric of trust and energy consumed for multi-hop routing to the Base Station. The protocol could reduce the error of an event's location as inferred by the Base Station in the presence of independently acting malicious nodes reporting false event locations. This work was published as a fast abstract in DSN 2005.
Equalization of 10 Gb Ethernet (GbE) multimode fiber links: This project was undertaken during my internship at Georgia Institute of Technology, where I was working on Optical Networks, specifically 10 Gigabit/s Ethernet using multimode fibers. I used the measured impulse response of multimode fibers of various lengths to simulate optical fiber links using Matlab. I studied the Inter-Symbol Interference (ISI) in the fiber, which is severe if left uncorrected. Various techniques based on equalization, namely Feed Forward Equalization (FFE) and Decision Feedback Equalization (DFE), were investigated to correct for ISI. I showed that these techniques work successfully and can extend the reach of 10 Gigabit Ethernet by 4x to 300m compared to the then IEEE standard, while maintaining the Bit Error Rate required by the standard. The results were published in IEEE LEOS Conference in 2003 and can be found here.
Design of IEEE 802.11g compliant Wireless LAN system: My undergraduate project was on simulation of IEEE 802.11g, which was the new IEEE Wireless LAN standard at the time offering bit rates upto 54 Mbps. My C++ simulations adhering to the specifications of the IEEE standard were used to evaluate the protocol. This helped the team investigate the performance of the standard in varying wireless channel conditions and was subsequently in the hardware implementation of the standard.
PhD in Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA
GPA: 4.00/4.00
MS in Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA.
GPA: 4.00/4.00
B. Tech (H), Electrical Engineering, IIT Kharagpur, India.
Minor: Electronics and Electrical Communication Engineering
GPA: 9.34/10.00, GPA in Minor: 9.54/10.00
Product management
Technical Leadership
Product Quality
Stakeholder management
Project management
Agile
Scrum
Product release
System Architecture
Cloud Computing
Data Mining and Machine Learning
Data Centers
Routing Protocols
Border Gateway Protocol (BGP)
Domain Name System (DNS)
Distributed Systems
Wired Networks
Ad-hoc, Sensor and Wireless Networks
P2P Networks
Security
Operations, Administration and Management (OAM)
FCAPS
Element and Network Management Systems (EMS/NMS)
Edward T.L. Hardie, David Craig, Ranjith S. Jayaram, and Ravish Khosla, "Methods and Apparatus for Distribution of IP Layer Routing Information in Peer-to-Peer Overlay Networks", Granted May 2014.
Ravish Khosla, "Improving Internet Infrastructure: BGP Predictability and Cloud DNS Performance", PhD Dissertation and CERIAS Tech Report 2011-15, Purdue University, December 2011.
Ravish Khosla, Sonia Fahmy and Y. Charlie Hu, "Content Retrieval using Cloud-based DNS", Global Internet Symposium, March 2012.
Ravish Khosla, Sonia Fahmy and Y. Charlie Hu, "BGP Molecules: Understanding and Predicting Prefix Failures", IEEE INFOCOM Mini-Conference, April 2011.
Ravish Khosla, Sonia Fahmy, Y. Charlie Hu and Jennifer Neville, "Prediction Models for Long-term Internet Prefix Availability", Elsevier Computer Networks, 55(3), October 2010.
Ravish Khosla, Sonia Fahmy, Y. Charlie Hu and Jennifer Neville, "Predicting Prefix Availability in the Internet", IEEE INFOCOM Mini-Conference, March 2010.
Ravish Khosla, Sonia Fahmy, and Y. Charlie Hu, "On the Impact of Filters in Analyzing Prefix Reachability in the Internet", International Conference on Computer Communications and Networks (ICCCN), August 2009.
Ravish Khosla, Xuan Zhong, Gunjan Khanna, Saurabh Bagchi and Edward J. Coyle, "Performance Comparison of SPIN based Push-Pull Protocols", IEEE Wireless Communications and Networking Conference (WCNC), March 2007.
Xuan Zhong, Ravish Khosla, Gunjan Khanna, Saurabh Bagchi and Edward J. Coyle, "Data-Centric Routing in Sensor Networks: Single-hop Broadcast or Multi-hop Unicast?", IEEE Vehicular Technology Conference (VTC), April 2007.
Ravish Khosla, "Reliable Data Dissemination in Energy Constrained Sensor Networks", Masters Thesis, Purdue University, August 2006.
Ravish Khosla, Yu Cheng, and Saurabh Bagchi, "Reliable Data Dissemination using Trust in Multi-hop Sensor Networks", Fast Abstract, The International Conference on Dependable Systems and Networks (DSN), June 2005.
Ravish Khosla, "Physical Layer Design of IEEE 802.11a compliant Wireless LAN", B.Tech Thesis, Indian Institute of Technology Kharagpur, May 2004.
Ravish Khosla, Kshitiz Kumar, Ketan M. Patel, Cattalen Pelard and Stephen E. Ralph, "Equalization of 10 GbE multimode fiber links", 16th Annual Meeting of the IEEE Lasers and Electro-Optics Society (LEOS), October 2003.
Awarded University wide Silver Medal for having the highest GPA in Electrical Engineering Department, IIT Kharagpur
Awarded Magoon Award for Teaching Excellence from ECE Department, Purdue University for receiving excellent reviews while teaching 400 students
Awarded Tellabs Research Assistantship at Purdue University for a period of 7 months
Awarded Graduate Teacher Certificate (GTC) by Center for Instructional Excellence, Purdue University
Voted the best student project in ECE 695N course on Wireless Networks at Purdue University
Awarded Bhartiya Cutler Hammer Prize and J.C. Ghosh Memorial Prize for outstanding academic performance at IIT Kharagpur
Received Certificate of Merit from Central Board of Secondary Education, New Delhi for scoring 99% in Mathematics and Physics at the Secondary Level
The Last Lecture by Randy Pausch
Sixth Sense, Integrating information with the real world
TED, Ideas Worth Spreading