Which service should you use?

A, DynamoDB.

While it’s very likely that you would want to use something like Kinesis to collect the votes from the mobile devices (this is a great use case for Kinesis), you still need a data store to drop the votes into and from which you can tabulate the votes. And the question is specifically asking about the data store, not about the means of collecting the incoming votes from the mobile devices.

Possible candidate services for the data store could be RDS, DynamoDB, RedShift, or ElastiCache. (The question asks about which service you would want to use for the data store, which eliminates running a distributed database or cache on EC2 instances.)

RDS is not a good fit because these would be incoming writes from the Kinesis stream. RDS is not very scalable from a write standpoint (other than building a much larger instance, which only gets you so far with a scenario of this scale), nor is it as low-latency as DynamoDB, and you need low-latency for “real-time” tabulation.

DynamoDB is a great option here, since it fits the requirements for durability, scalability, and high availability, and can operate with low-latency to meet the “real-time” tabulation requirement. It can easily scale out to handle the large amount of incoming writes to increment the vote values for candidate names in a table as you pull data out of the Kinesis stream.

RedShift is probably not a ideal fit, since you’re not looking to do complex analytics across many records or tables taking up terebytes/petabytes of data here, you’re looking to just take incoming votes and increment the vote values for your candidate names as they come in. A few records in a single table should suffice to capture the max dozen or so possible TV show candidates voters would be choosing from. (I wouldn’t want to watch a TV show with so many candidates that you would need millions of records to track all of their names!). So while Kinesis Firehose ingesting to RedShift is an option here, it’s not as ideal an option as using DynamoDB for the data store.

ElastiCache would definitely be low latency, and could even do the vote tabulation for you in real time upon data insertion into the cache if you used Redis with sorted data sets. And selecting Redis would also get you the durability option with backups to S3, which you couldn’t get with Memcached. Really the only knock on ElastiCache running Redis for this scenario would be that you can’t scale it horizontally for writes with sorted data sets, which eliminates it as an option for something of this scale. ElastiCache with Memcached might meet the scalability requirements because you could shard it horizontially for writes, but Memcached doesn’t have the option to back up its data to S3 like Redis does, so it wouldn’t meet the durability requirement. If you lose a Memcached node, the data on it is gone.

So for the possible data stores, RDS, RedShift, and ElastiCache are out. That just leaves DynamoDB. Which is a great candidate, and is one of the possible answers.