This autumn I signed up for a programming competition in efficiency hosted by Consid.
In Sweden a car’s license plate has three letters followed by three numbers, such as
ABC123. The task was to as quickly as possible answer whether or not a duplicate license plate existed in a given textfile of 6 million entries. I didn’t win, but managed to get pretty close with completion time within 50ms. My solution can be found here, but I wanted to explain my thoughtprocess a bit here.
In order to scale the project with multiple cores I wanted to include parallel processing. However, reading and writing to/from a file is done sequentially. So the first thing that is done is to open the given textfile and save it to a dynamically allocated buffer. By doing this can I quickly access different parts (or license plates) from the textfiles in parallel without compromising thread safety.
Since the data can be handled in parallel I needed an efficient way to store information of the status of a license plate number. To do this I created a boolean matrix with
1000columns. I then translated the license plate string into a character part,
c, and number part,
n. The number part corresponds for what column in the matrix the license plate belongs to and the character part to which row. However, since the character part contains letters I needed to convert this to a unique integer ID instead. This is done by converting
cto base-26 index, letting
A = 1, B = 2, ..., BA = 2 * 26 + 1.
Now can the license plate number be converted to a unique cell in the boolean matrix with id
[c][n]. That location is then set to store
nare private variables for each working thread and there’s no discontinuity in the shared matrix variable. Once all of the data has been read through can a parallel loop with a reduction clause be used to find the number of stored
true-variables in the matrix. If this value differs from the number of read license plates, then a duplicate exists. (
lSizeis the number of characters in the data,
lSize/8is used to count number of license plates in the file).