|
> On the other hand, the ordering T_0,A,0 < T_0,A,1 < T_0,B,0 < T_0,B,1 is arbitrary. Still don't see it. We are already making the assumption that the millisecond part is the same T_0. So then seq will depend strictly on the count of previous such IDs issued by that node only (would you agree with that?). So that is why I said that it doesn't make sense to compare seq number from A -- (0) and a seq number of B -- (1) before considering the identity of A & B. You would effectively order by relative load between all your machines at that particular time, at that time frame it would be things like disk access, cache states and so on. EDIT: In addition. Ordering T_0,A,0 < T_0,A,1 < T_0,B,0 < T_0,B,1 will be stable though, machine A,B mac addresses don't change. For any particular millisecond you will get all the results for a particular machine then those results will be sorted by sequence # so you would even get some kind of a relative load measurement. Let's look at a longer example: T_0,A,0 T_0,A,1 T_0,B,0 T_0,B,1 T_0,B,2 T_0,B,3 T_0,C,1 T_0,C,2 T_0,C,3 T_0,D,0 I think is a lot better order than T_0,0,A T_0,0,B T_0,0,C T_0,0,D T_0,1,A T_0,1,B T_0,1,C T_0,2,B, T_0,2,C T_0,3,B T_0,3,C |