There are two elements you need to provision when creating your Amazon MSK clusters: Broker instances and Broker storage. A Broker instance is a worker node that helps to manage the Kafka cluster. Your clusters can have multiple brokers, but can also operate as a single node, if necessary. Broker instances can be run within the same availability zone or across many availability zones to create a highly available cluster, something that many architectures will require.

One of the large benefits of using Amazon MSK over a roll-your-own version of Kafka is that Amazon will keep an eye on these broker nodes and replace them if they become unhealthy. Otherwise, you would have to manage all of that yourself. Broker storage is where all of your data will be kept that comes streaming into Amazon MSK.

Within Amazon, this storage is housed within EBS volumes, and gains all the protections that EBS provides, like durability and fault tolerance. You can also encrypt these data volumes using Amazon EBS Server-side encryption and AWS KMS, the Key Management Service. An interesting note about storage however, is that once you have assigned your Broker storage, you can only increase the amount of storage, it can not be decreased. The broker's storage can be autoscaled upwards if you do reach your maximum capacity.

If you already have a Kafka cluster that you are managing yourself, either on-premises or within the cloud, you can migrate over to Amazon MSK. There are many third-party and open-source tools like MirrorMaker which can help replicate the data from your Kafka cluster into Amazon MSK.