Improve performance of rename class functionality

[andrii0lomakin]

During serialization of document we encode name of class of document as part of binary content. As result if we rename class and have big amount of data in database it becomes very very slow.

I can see following options:

1. Set by convention that we can not mix records from different classes in one cluster (not so big problem I think) and do not encode class name in record it will be detected automatically. As result we will not have problems with rename and overhead of such detection can be very minimal.
2. Set unique identifier for each class and use it for serialization/deserialilzation. This approach is more flexible and robust.

So we need to choose something ))

[lvca]

At the beginning the idea of layered independent blocks was sexy, but after few years how many people mix different records in the same cluster? I think you're right. This would reduce the record size, improve serialization performance and other operations like class rename. So +1 for me. The perfect place is 2.0 because it's a huge change.

[andrii0lomakin]

Some users use single cluster for several classes approach. Here is use case https://groups.google.com/forum/#!topic/orient-database/ih6tOvL7w7A

[lvca]

We could remove this feature and oblige users to use OO inheritance by creating multiple classes ;-)