(stolen out of Mikael’s post to the cluster list – http://lists.mysql.com/cluster/2396)

Both REDO and UNDO logs are stored in buffers but are sent after a while to disk.

For REDO the disk write happens when 256 kByte of buffer is filled up or when a GCP requests synching the data (the 256 kByte write is without synch
but every 4 MByte gets synched even without GCP occurring.

For UNDO similar algorithm that write when 256 kByte of buffer is filled, synch after 4 MByte or when the local checkpoint is completed.

I wonder if we should be tuning more? I honestly don’t know the answer to this – but I don’t think it’s a limiting factor at the moment :)