Comments for post Sorting in key-value data model
800 numbers writes: I must say that the key value data model is catching attention in a pretty large scale recently and I think that their ability to represent data in the mode of a map between keys and values very efficiently is what that makes them more applicable and practicable. What is interesting about the key value data model is that it is new practically but old theoretically, which thereby makes this the end choice for many users!
antirez writes: @Matteo: thanks for your comment. For now we have a single ordering weight vector, the idea is that if you have more than one, there should be a client that will take care to sync into an unique weight all the other indexes.
About the problem of N servers, your method is workable of course, but I think that key tags are a more general way of making sure a set of N keys will reach the same server.
Matteo Merli writes: Hi, this approach for sorted items is really interesting. Reading this I would think of a system in which when you insert a key you can specify an index (or more than one) like:
put( Key, Value, [MyIdx1=X, MyIdx2=Y] )
this way the (Key,Value) will be inserted with usual hashing, and then the key will be added also to the specified indexes. Now every index will be routed (as a key) with hashing to the corresponding server.
then I could have something as :
get_range( MyIdx1, 1, 50 ) -> [K1, K2, ...]
that will involve the query only to the server that owns MyIdx1. Anyway, this doesn't solve the problem when selecting using more indices, as the could resides on different servers.
home