Do we really need The Globally Unique IDs?
Posted: Tue Jan 29, 2013 7:22 am
Collegues, it seems that we have a problem!
1. We (Rosatom RDS developers) really don't understand what profit in usage of Globally Unique IDs?
Because old-style URIs = namespace + LocalID (no matter human-readable or not) will be globally Unique by definition (even if local ID is just a sequence number). So what we need in Globally Unique IDs anyway? What problems they are solving?
2. Moreover, R-UUID things using without namespaces brings a really serious performance issues:
When we are walking throught the graph that consist of nodes from different RDLs we must to poll all available RDLs just to understand from what namespace this R-UUID from for every (!) single node. Some ideas of usage of "master look-up table" in core RDL is a performance penalty too. So a word from technical guy: we don't use an own network protocol or stack. Its all about HTTP. So we don't need a MAC-like R-UUIDs and DNS-like "master look-up tables". We are already have DNS, thank you.
So we really want to understand for sake of what technological advantages we are sacrifying technical consistency. All ideas will be highly appreciated!.
Nice example: try to imagine that all names of towns, streets, houses were abolished and invented a global registry of persons, so if you want to know where exact person lives try to call to information desk of all available towns until desired person will be founded. Some sort of bizzare isn't it?
1. We (Rosatom RDS developers) really don't understand what profit in usage of Globally Unique IDs?
Because old-style URIs = namespace + LocalID (no matter human-readable or not) will be globally Unique by definition (even if local ID is just a sequence number). So what we need in Globally Unique IDs anyway? What problems they are solving?
2. Moreover, R-UUID things using without namespaces brings a really serious performance issues:
When we are walking throught the graph that consist of nodes from different RDLs we must to poll all available RDLs just to understand from what namespace this R-UUID from for every (!) single node. Some ideas of usage of "master look-up table" in core RDL is a performance penalty too. So a word from technical guy: we don't use an own network protocol or stack. Its all about HTTP. So we don't need a MAC-like R-UUIDs and DNS-like "master look-up tables". We are already have DNS, thank you.
So we really want to understand for sake of what technological advantages we are sacrifying technical consistency. All ideas will be highly appreciated!.
Nice example: try to imagine that all names of towns, streets, houses were abolished and invented a global registry of persons, so if you want to know where exact person lives try to call to information desk of all available towns until desired person will be founded. Some sort of bizzare isn't it?