How come lazy versioning has no fault tolerance issues? Can't it also crash in the middle of flushing its write buffer? I suppose this is less likely since simply writing the data will take less time than performing the operations in a transaction, but either way we are reliant on proper log keeping to make sure that we can detect if it happens and to abort the changes made if it does.
This comment was marked helpful 0 times.
mchoquet
@nrchu I think the idea is that the write buffer flushing is done in hardware, so we can check it ahead of time and know it will never fail. Eager versioning has a fault tolerance problem not because its log keeping infrastructure is more complicated, but because it has to run untrusted code (i.e. your code), and if your code fails then it needs to go fix memory.
How come lazy versioning has no fault tolerance issues? Can't it also crash in the middle of flushing its write buffer? I suppose this is less likely since simply writing the data will take less time than performing the operations in a transaction, but either way we are reliant on proper log keeping to make sure that we can detect if it happens and to abort the changes made if it does.
This comment was marked helpful 0 times.
@nrchu I think the idea is that the write buffer flushing is done in hardware, so we can check it ahead of time and know it will never fail. Eager versioning has a fault tolerance problem not because its log keeping infrastructure is more complicated, but because it has to run untrusted code (i.e. your code), and if your code fails then it needs to go fix memory.
This comment was marked helpful 0 times.