The Problem: The style of the contest was quite familiar - a given problem and associated input formats, guarantees, etc. The problem itself dealt with finding the shortest one-way itinerary between the given origin and destination airports. The knowledge base is a list of all flights between the ~100 airports, each entry accompanied by its timings and unique flight numbers. The required output is a set of 5 best itineraries where the comparison metric was (in that order) least total duration, least layovers (cities on the way) and smallest flight numbers(?). We were provided a small sample database of flights and a handful of sample queries and their correct responses for testing, and the actual magnitude of the real tests were unknown. Thus, one of the main motivations was to tighten the solution as much as possible. And finally, the allowed programming languages were Python, Ruby and Java.

My solution: I chose to implement this as a Djikstra-like graph shortest path solution. Hence intuitively each city (airport) would be a node in the graph and flights would provide the directed edges between them. This can be implemented using a priority queue (standard idea). However, an interesting aspect of this particular problem is that not all incoming flights at an airport can be paired with every outgoing flight. This is because of their actual arrival and departure times respectively which limits the possible pairings. Therefore a flight arriving at time t at an airport can only be followed by flights departing at time greater than t at the same airport (the minimum layover time was defined to be 1 minute).

Although this means that the total number of possible paths between the origin and destination are limited, the search had to be context driven and it is not possible to discard airports as useless using a single flight path. It could definitely be possible that a flight path through an airport have a very high cost, but a flight that reaches earlier (but discovered later) could very well have a better cost. I settled for maintaining the whole flight path in every search path and contextually identify matching flights. As in Djikstra's solution, these paths were still being maintained in a priority queue and only the best were picked at each iteration for inspection. I did give this some extra thought for optimization possibilities such as a maximum limit on the best paths that would be maintained for each airport, but its required code looked cumbersome and inefficient, and moreover hard to prove. My implementation was in Python (as I was already familiar with it, and I definitely prefer it over Java!). Each flight had its own class object and hence the itinerary object just had to keep pushing the flights into a list. The itinerary object eventually went into the priority queue.

Optimizations: Of course, I optimized everything out of the critical loop handling node expansions. I obtained the most performance gains from profiling my code using the inbuilt Python profiler. I realized immediately that the itinerary object was too big and its deepcopy (during expansion at a node) was consuming the most time in a run. Removing the class data fields one by one and replacing their usage with method calls, I realized that recomputing them was actually faster than storing and copying them everywhere. The final fastest implementation I obtained was probably not what I would have thought of as good looking code, but that was what I wanted! (forget Software Engineering temporarily). In total, I reduced the time by almost half.

Winners were supposed to be notified sometime in November last year, and hence I conveniently assumed that I had not won, and I was not able to find any related news articles on the web either. I got an email today morning from Orbitz recruitment that I had won along with my free round trip voucher! I am still unaware of other winners and hope to hear some statistics! And finally, I hope to use this to travel to some new place that provides excellent opportunities for photography and something new&different.

I don't want to post the solution here because of complex legal agreements, but if you are really interested, send me an email (nkarthiks@gmail.com) and we can talk.

• The official Orbitz Coding Challenge 2010 webpage (Google doc)

It was fun working with KC, Luke and Patrick on this paper, which was the outcome of a Distributed systems course project. We were able to build a system to show significant network performance gains over naively programmed Java programs. These gains were the benefit of two simple techniques known as Batching and Chaining. Our main contribution was the automatic code transformation of normal Java code into network-efficient code using Session types.

Also this year, our work on 'Finding Latent Performance Bugs in Systems Implementations' was accepted for publication into FSE 2010. It was three rewards in 6 months!

The PDF version of the original paper published at Coordination 2010.

The research which was secretive for a long while was unveiled last week, led by KR Sreedhar. He happens to be an alumni of NIT, Trichy: a proud moment for all us Nittians! (His profile says Madras University but it happens so that REC, Trichy back then was not autonomous, but was affiliated to Madras University).

Catch the following video to get a first hand update on whats new about Bloom energy fuel cells.

Bloom energy on CBS

So once I was partially convinced of buying new RAM, I went over to Apple Store and saw that 4GB of compatible RAM costs $200! That was a bit too much for just RAM. A little browsing around told me that Apple made customers think that Macbooks are too picky with RAM, and may not work with all memory or brands.

More browsing with sellers like MacConnection and MacMall gave me utter confusions about memory specs and compatibility. I wish I had come across MacRumours guide to buying new RAM earlier. So I followed their advice and downloaded the Crucial Memory scanner to identify my system model and the right memory. I had the 2.4GHz 13" late 2008 aluminum unibody model (MB467LL/A) and crucial found me the right RAM with some model number (CT898327).

Using more Google in parallel, I arrived at another popular Crucial model number (CT2KIT25664BC1067) which had worked for people before. And with the final search with both these model numbers, I landed on a forum where I learned that someone had inquired with Crucial itself, and these model numbers were indeed the same (marketing tactics!).

Finally, the Amazon search led me to this, and I was able to buy it for $20 cheaper than Crucial's website, and $100 cheaper than Apple's website! Happy as I was, the next challenge was to get the RAM fitted properly into my mac. Again, Mac rumors pointed me to OWC's website which has awesome installation videos.

If you have the right type of screw drivers and follow the instructions, it indeed takes less than 10 minutes to get the job done and boot up into 4Gigs. Well, its a positive experience and I've not had any issues for 2 days Now, I can run Windows 7 and see how good it is really. And seriously, Microsoft has done a good job in making it look good and yet feel as light as XP. It has been a year since I left Windows completely and I have not had any issues that make me want to use Windows - I am a happy Apple customer.

Specs: 4GB Kit (2GBx2), 204-pin, PC3-8500, SODIMM, DDR3, Non-ECC, Unbuffered. Amazon, Crucial, Apple store.

I have been keeping myself busy primarily with my PhD coursework in my 4th semester at Purdue. This is the time when I finish most of my coursework and delve into research. I'll be taking my final oral Qualifier exams this semester and I hope it goes well and I officially become a PhD candidate. We worked on a couple of paper submissions and the reviews are pending on them.

Last Thanksgiving tempted me a lot into buying my first DSLR camera - a Canon 500D (Rebel T1i). I can see people going - "Aah, one more guy with a fancy camera", but I really hope I can impress myself with my photos. I truly believe that photos keep memories alive through the years. Anyway, I've decided to start with the kit lens 18-55mm and a 55-250mm telephoto zoom lens. I've transformed into a Flickr user over Picasa after receiving recommendations from a couple of friends and here is my Photostream. The zoom lens is producing some superb images when I use it for shooting portraits with easy background blur.

My camera is my second best companion while traveling. I prefer spending and visiting different places instead of lazing back at home. Last winter, I went on a small road trip with Ashwathi to Brown County state park. I was pessimistic at first about the worth of this trip during winter, but it turned out to be quite a nice trip. Brown county park is located near the small town of Nashville - I really mean small; the whole town is built around Van Buren street. The park was a quiet escape and the weather was in the 40s and partially cloudy with bright skies, perfect for photography. I'd love to head out to a similar park sometime in Fall to camp and spend more time close to nature. And finally, I spent a quiet and exciting New Year back home at W Lafayette with an Italian Dinner at Olive Garden.

I've also been doing this one-new-restaurant-every-Friday thing, which actually started out last Fall. A couple of us guys pick some new cuisine and restaurant every Friday and try it out. Although we've had a few experiences that were not too good, we get to try out so many different styles of cooking. And after all, its TGIF!

Meanwhile, I also caught up on movies such as Sherlock Holmes, 3 Idiots (Hindi) and Avatar (3D IMAX) and a couple of other older movies. I am so happy we discovered the Purdue DC++ network for content. We had a similar one in our GZ LAN at NITT and this brings memories back..

Back home again! #

@suren Lets plan to meet. You can contact me on another person's number (if you don have my old number) #

Powered by Twitter Tools.

Back home again! #

@suren Lets plan to meet. You can contact me on another person's number (if you don have my old number) #

Powered by Twitter Tools.

@divinjohn IIT, Madras is not on the list! #

Going back home after 8 months! excited #

Happy Mothers Day to all mothers out there! #

Powered by Twitter Tools.

@verma Poster went well. Got a lot of feedback and motivation to work on it #

@rvivek Leaving home or leaving for home? Sounds like u are running away from home (with someone? ) #

@satya3656 Introducing yourself to twitter? #

Youtube has a better UI! Nice #

@viveknshah To get some overhead baggage space #

Powered by Twitter Tools.