We are configuring Oracle 10.2.0.4 2-node RAC database on Linux x86_64

I was going through Oracle documentation and a few books which were discussing about cache fusion (read-read, read-write, write-write, write-read behaviour) etc.

It was mentioned that row locking is similar to that of single instance so whevever a block is needed a instance completes its operation on the block and transfers the block to the other instance which needs exclusive access.
What I understood is once the transaction is completed the first instance transfers the block to the other instance to have exclusive access to that block.

Also the first instance retains a PI(Past Image) of the block which it can use for read operations but it cannot modify the block as it has transferred the block to other instance and it only has shared access.

Now when a checkpoint occurs all the instances should purge their PI's so in that case if the first instance is in the middle of a select statement which uses the PI of the block what will heppen to the query, will it result in an error(something like snapshot too old) ?

I am not able to understand the PI concept and what happens during flushing the data blocks to the disk.

Does it mean multiversion read consistency doesn't have anything to do with PI ?

Multiversion read consistency is a part of even a single instance database so what is the use of having a PI in RAC ?

Since a specific instance has done some changes and then has transmitted the block which means a new SCN would have been created which in turn means multiple versions of the block exists which is similar to single instance database.

So what I don't understand is what is the significance of the Past Image here ?

Thanks Jonathan.

I didn't think from recovery point of view actually I was thinking with respect to cache fusion perspective.