OK. Before I start, one careless error crept in my earlier post (since edited), that data is written to datafile on checkpoint/commit. Commit has got nothing to do with writing data from db buffer to datafile. (Although it does cause LGWR to write data from log buffer to online redo logs). DBWR process (es) write dirty buffers to disk on checkpoint/ timeout or when the number of dirty buffers cross a threshhold.
Originally posted by dbafreak:
When the server crashes or say during a power shut down, assuming ONLINE Backup -- Can we get the uncommitted transactions from the redolog files? I believe that redo logs only have committed transactions. In which case, whats the difference between the backed-up DATA FILES and backed up REDO LOG FILES? Please explain

Uncommitted transactions will be lost.
Redolog files have all transactions which have been written to online redologs by LGWR (Both committed and uncommitted, less those still in log buffer), stored transaction-wise along with their SCN.
Datafiles have the data upto the last Checkpoint. In case of a crash, and subsequent startup, SMON detects that instance recovery is required because checkpoint sequence numbers in datafile headers are not synchronised with online redologs.redo is applied first from the redo logs (roll forward) followed by rollback upto last committed transaction.