First of all, Kafka, AFAIK, is not a storage system , but rather a messaging platform that you can use to send time series events to a system that will process and store this information. p>
Secondly, it all depends on how you want to aggregate, query, and retrieve data. As with most NoSQL databases, these issues are crucial before you even think about moving forward.
Thirdly, there are two solutions for storing time series data built on top of both HBase and Kassandra, namely OpenTSDB and KairosDB . Both are fairly well known, both should be able to store events with high resolution (seconds or milliseconds). Or, alternatively, you can take a look at Druid , but this is a slightly different approach to the problem ...
In general, I would suggest KairosDB - it turned out to be an excellent performer for storing TB data of time series.