When cluster is in quorum, nodes can communicate properly with each other, and cluster level things are handled properly.
Then you have a master node, which is voted by nodes in the cluster.
Master node makes sure that data is synchronized between cluster nodes.
If master node goes down, new master is selected by remaining nodes by voting.
If cluster loses quorum for some reason, this voting thing comes in again, each node has one vote.
If you have 6 node cluster, one ha pair in one room, and behind a wall are two ha pairs and for some reason cluster network goes down between these two rooms, room where's 4 nodes continues serving data, and this lonely ha pair stops serving data. It's a voting thing, two nodes against four.
If you have two ha pair cluster, one ha pair in one room, and second ha in another, vote result would be a tie.
epsilon is exta vote for a node which is holding it, by that way new master is voted, and there is no tie. Quorum would be handled by ha pair which is holding the epsilon in this case.
2 nodes without epsilon wouldn't serve any data, and 2 nodes with epsilon would hold the quorum and continue serving data.
Epsilon is not reassigned if the node which is holding it goes down, if it's a master node at the same time, new master is voted by remaining nodes.
"From the above output if cluster1-n1 and cluster1-n2 goes down. Is there any way to access aggregate owned by these nodes?" Normal takeover/giveback things are still existing in c-mode, if both nodes goes down in a ha pair you cannot access to shelves behind these nodes.