The SSTables Per Read option displays the latest mean and maximum metrics for each column family, averaged across the cluster.
The Mean column represents, on average (across all nodes over roughly the last five minutes), how many SSTables must be involved per read request.
The Maximum column represents the highest number of SSTables (across all nodes over roughly the last five minutes) that were involved in a read request.
High numbers of sstables per read (typically, more than 3 or 4 as a guide) can reduce read performance and if read performance is below desired levels you many need to change compaction strategy for the affected column family.
The SSTables Per Read option shows the mean and maximum SSTables involved per read request per node, over the last hour.