Verifying, identifying and ack-ing requests ....

[Oliver Spatscheck <spatsch@research.att.com>]

Abhi Deshmukh writes:
 > I have a set of questions on Identification, Verification and
 > Acknowledgement of requests, that would help me understand the current or
 > proposed working of peered CDNs. Here goes.
 > 
 > Assume that CDN-A and CDN-B peer with each other. 
 > Assume that CDN-A forwards an end-user request R to CDN-B.
 > 
 > 0. How does CDN-A know that CDN-B has succesfully (or un-successfully) 
 > received the request R ?

I don't think it is a requirement for CDN-A to know this. If for example DNS
based redirection is used CDN-A will never know if a end-user got a response
(with or without CDN peering). So we don't make things worse in that case.

On the other hand CDN-A is most likely interested in knowing if CDN-B
fulfills its SLA (might guarantee a 99.999% redirection success rate),
however, I thought that we wanted to keep SLA enforcement outside
the scope of this working group.

 > 
 > 1. How does CDN-A know that CDN-B has succesfully (or un-successfully) 
 > serviced the request R ?
 > 

I think one important aspect here is that one redirection might result in many
serviced requests. So we have R which is a request routing request ( for
example, DNS resolution) and S which is a service request (for example, HTTP
GET) and in general there is a one to many mapping between R and S. I also
think my comments to 0. apply here.

 > 2. Does CDN-B send a real-time or batched response back to CDN-A after the
 > request R is serviced ?
 > 

As pointed out in the request routing requirements draft we will need
some aggregated pseudo real time feedback. However, I feel the actual
detailed logs can be batched.

 > 3. How does CDN-A verify the work (including QoS, SLA, etc) done by CDN-B
 > while trying to service request R ?  
 > 	3.a  Do CDN-A and CDN-B share common/neutral monitoring apps at each
 > service points ?
 > 

See comments above. Is the verification part of our requirements? I would
like to know how other people think about this.

 > 4. Is each request between CDN-A and CDN-B identified uniquely ? Is it as
 > simple as a combination of timestamp and a rolling counter ?
 > 

The request routing request or the service request?


Oliver