Agreed. Ben

On 9 Jul 2013, at 22:41, "Y. Richard Yang" <[email protected]> wrote:

> 
> On Tue, Jul 9, 2013 at 5:22 PM, Ben Niven-Jenkins <[email protected]> 
> wrote:
>> 
>> [BN] Even if a single server is calculating both the network & cost maps 
>> then you are likely to have multiple servers for resiliency or due to demand 
>> load and might put them behind a load balancer. Because the network map and 
>> cost map are retrieved via separate requests you cannot guarantee both 
>> requests hit the same backend server (e.g. Because a server may fail or be 
>> removed from the load balancer pool) so you need to be able to 
>> deterministically generate vtags across different servers even for your use 
>> case I think. 
> 
> Agree that different requests can hit different servers, and I interpret 
> "deterministically" as the following:
> 
> two network maps have the same vtag => they are the same map
> 
> Note that the preceding condition allows some usage of random variables; for 
> example, in the case of only a master who computes the vtag, and the other 
> load balancers are just data/storage servers, then the master can use random 
> variables, or nonce...
> 
> Of course, the preceding is internal implementation. We just need to specify 
> the semantics, as the preceding "deterministically" mean. Agree?
> 
> Richard
> 
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to