Cluster and Cloud Computing Assignment 1 – Social Media Analytics
The suburb here is Abbotsbury and this is part of Greater Sydney (indicated by 1gsyd above). You will find that some suburb names (strings) are repeated many times, e.g., the suburb name Richmond occurs several times in New South Wales, Victoria, South Australia, Tasmania etc and also appears as substrings of suburb names, e.g., North Richmond,Richmond Lowlands, Broad Water (Richmond Valley – NSW) and some of these are in Greater Capital cities and rural locations. Your solution should aim to tackle such issues and dealing with potential ambiguities and/or flag cases where tweets cannot be resolved to a single unique location.
The tweets themselves can include the precise location, e.g., Kempsey, New South Wales, or the approximate location of where the tweet was made, e.g., Australia or New South Wales or indeed potentially non-Australian locations. Such approximate and non-Australian locations can be ignored from the analysis, i.e., unless the specific string is in the sal.json file it can be ignored. You can assume that city names given in tweets such as Sydney, New South Wales corresponds to Sydney (13730), Melbourne, Victoria corresponds to Melbourne (21640), Brisbane, Queensland corresponds to Brisbane (30379) etc.
Each tweet also contains a unique author id for the tweeter, e.g., 836119507173154816 for the first tweet in the tinyTwitter.json file.
Task 1: Your solution should count the number of tweets made by the same individual based on the bigTwitter.json file and return the top 10 tweeters in terms of the number of tweets made irrespective of where they tweeted. The result will be of the form (where the author Ids and tweet numbers are representative).
Rank Author Id Number of Tweets Made
Task 2: Using the bigTwitter.json and the sal.json file you will then count the number of tweets made in the various capital cities by all users. The result will be a table of the form (where the numbers are representative).
For this task, you may ignore tweets made by users in rural locations, e.g., 1rnsw (Rural New South Wales), 1rvic (Rural Victoria) etc.
Task 3: Finally, your solution should identify those tweeters that have tweeted in the most Greater Capital cities and the number of times they have tweeted from those locations. The top 10 tweeters making tweets from the most different locations should be returned and if there are equal number of locations, then these should be ranked by the number of tweets made as shown in the table below – again with representative data.
Rank Author Id Number of Unique City Locations and #Tweets
Note that for this task, only those tweets made in Greater Capital cities should be counted, e.g., if author Id = 5678910111213141516 tweets 1000 times from rural New South Wales then these can be ignored.
Your application should allow a given number of nodes and cores to be utilized. Specifically, your application should be run once to search the bigTwitter.json file on each of the following resources:
1 node and 1 core;
1 node and 8 cores;
2 nodes and 8 cores (with 4 cores per node).
The resources should be set when submitting the search application with the appropriate SLURM options. Note that you should run a single SLURM job three separate times on each of the resources given here, i.e. you should not need to run the same job 3 times on 1 node 1 core for example to benchmark the application. (This is a shared facility and this many COMP90024 students will consume a lot of resources!).
You can implement your solution using any routines and libraries you wish however it is strongly recommended that you follow the guidelines provided on access and use of the SPARTAN cluster. Do not for example think that the job scheduler/SPARTAN automatically parallelizes your code – it doesn’t! You may wish to use the pre-existing MPI libraries that have been installed for C, C++ or Python, e.g., mpi4py. You should feel free to make use of the Internet to identify which JSON processing libraries you might use. You may also use any regular expression libraries that you might need for string comparison.
Your application should return the final results and the time to run the job itself, i.e. the time for the first job starting on a given SPARTAN node to the time the last job completes. You may ignore the queuing time. The focus of this assignment is not to optimize the application to run faster, but to learn about HPC and how basic benchmarking of applications on a HPC facility can be achieved and the lessons learned in doing this on a shared resource.
Final packaging and delivery
You should write a brief report on the application – no more than 4 pages!, outlining how it can be invoked, i.e. it should include the scripts used for submitting the job to SPARTAN, the approach you took to parallelize your code,and describe variations in its performance on different numbers of nodes and cores. Your report should include the actual results tables as outlined above and a single graph (e.g., a bar chart) showing the time for execution of your solution on 1 node with 1 core, on 1 node with 8 cores and on 2 nodes with 8 cores.