As processing can be burdensome for the relational database, you often have to come up with ways that avoid reading all the information again and again from the sources. One option for avoiding this is the use of partitions. Just like in relational tables, partitions are a way to divide a large table into smaller chunks.
Imagine that you have a large fact table that contains all the order rows, every time that you process your measure group, you will read the entire table. This is probably not necessary as you do not change old order rows on orders that have been fulfilled. You will only add new order rows. If this is the case, then you can work with partitions. For this, you need to identify a good partition key. A good partition key will vary, but you need to ensure that older partitions are not updated. If they do, you would need to reprocess the entire cube.
You may want to have a number of partitions but not too many; the administration should...