Created by: jetfuel

At the moment, each PersistToDisk function call will trigger proto.SerializeToOstream(&file); and its a very expansive operation. The system is rewriting the whole data to the filesystem. This is inefficient and the operation will get more expansive as the data grows.

We should research if Protobuf allows us to append data into an existing file. That should allow VisualDL to work with training that takes huge data points.