Joel, What you stated is OK. Maybe you can soften requirement on I2RS agent to remember everything after restart. Many things can happen while the agent is down. The information in the agent's non-volatile DB may be stale.
Can authors address the other issues I raised in my original comments? Linda -----Original Message----- From: Joel M. Halpern [mailto:[email protected]] Sent: Wednesday, February 26, 2014 10:45 AM To: Linda Dunbar; [email protected] Subject: Re: [i2rs] Comments to draft-ietf-i2rs-architecture-02.txt Since 1) There is no requirement for such a coordinator. (The fact that I personally expect it to be rare is irrelevant.) 2) Such a coordinator if used is not known a-priori to the I2RS Agents I do not see that the I2RS Agent behvior can make use of this to simplify the restart notification behavior. I would love to simplify that. But this does not seem to do what is needed. On 2/26/14, 11:35 AM, Linda Dunbar wrote: > Joel, > > Agree that "I2RS Clients locate I2RS agents by many means". > For each method, there must be a coordinator (or system) to facilitate the > association. My main point is let the "restart status" being sent to this > "coordinator", instead depending on the I2RS agent's non-volatile DB. It is > possible during the time that this I2RS agent is down, the client is > reassigned to a different agent. > > Linda > > -----Original Message----- > From: Joel M. Halpern [mailto:[email protected]] > Sent: Wednesday, February 26, 2014 10:23 AM > To: Linda Dunbar; [email protected] > Subject: Re: [i2rs] Comments to draft-ietf-i2rs-architecture-02.txt > > Linda, (think you for trimming.) > > On 2/26/14, 11:17 AM, Linda Dunbar wrote: >> Joel, >> >> --- Joel Halpern wrote --- >>> I2RS clients do not find I2RS Agents. I2RS Agents are contacted by >>> I2RS Clients. The closest they come to "finding" is if the I2RS >>> Agent registers for notifications, then the I2RS Client will later >>> send that I2RS Agent notifications. But that is dynamic information >>> created by the I2RS Agent. >> >> [Linda] I meant to ask "at the very beginning, where does the I2RS >> clients get the information on which I2RS agent to register for >> notification"? Is it by some management entity? If yes, then when the >> I2RS agent restarts, the I2RS agent should send a notification to >> this "management entity", instead depending on the I2RS agent's >> non-volatile DB. It is possible during the time that this I2RS agent >> is down, the client is reassigned to a different agent. > > I2RS Clients locate I2RS agents by many means. We are not mandating them. > For example, an I2RS agent may start by using local means to find its local > router. It can then use I2RS mechanisms to get the topology. > Based on that, and the service it wants to provide, it can then > select which I2RS agents to register with and what events to register > for. But that is only one example. It may be that some other entity > (a PCE for > example) provides the needed topology information. Or any number of other > deployment models. The I2RS Agent failure notification mechanism ought not > be dependent upon the I2RS client deployment and management strategy. > > Yours, > Joel > >> >> >> >> Linda >> > _______________________________________________ i2rs mailing list [email protected] https://www.ietf.org/mailman/listinfo/i2rs
