Hi Mike,

Thanks for reviewing our proposal and providing your considerations. 

My comments inline below. 

Thanks,

Peter Lee

On Oct 17, 2014, at 6:42 AM, "Mike Young" <myoung@...> wrote:

Yes we are including storage reservations as one of the functional deliverables of this project. 

Our initial approach to resource reservation will be utilizing the currently available quota extensions on Nova, Cinder, and Neutron.  Essentially, the quota limits on a per project/tenant will represent available capacity across each of the infrastructure resource elements configured into a resource pool. 

There will not be "pre-allocation" of resources since the reservation system will not be shelving VMs or such but instead applying time based capacity availability models to make decisions regarding future availability of resource capacities. 

As to storage, we will be modeling based on gigabytes, snapshots, and volumes.  Any additional resource elements will require upstream contribution to Cinder to introduce additional quota elements. 

We are currently sticking with block based but open to consider other models. 

Currently, the actual allocation/provisioning of such resources is out-of-scope of the resource reservation service, the reservation service does not implicitly make or provision or allocate the instances (VMs or volumes, etc) but rather the higher layer NFVO or VNFM would actually perform their needed allocation operations based on a successful reservation ID to begin consuming them according to the start-time defined for the reservation. 

Basically, the reservation service informs that reserved resources have been preserved for use (say 100GB between X time and Y time) and the consumer of that reservation will decide how they go about using it (say 3 volumes each 25GB) if and when they decide to use it within the specified reservation validity period. 

I think these are important considerations for the actual provisioner of reserved resources to ensure timely availability operations can be ensured when they need it.  We can think of reservation service as providing "check-in" time and "check-out" time (like booking a hotel room) but the person needing to use the actual reserved room should be mindful of how long it will take to get to the room and use it for their needs.

For example:

If the actual volumes need to be created, filled with snapshot data, and brought online for use by a given VM by X time, we can consider the reservation service request to specify the start time as X - 10 mins and the requestor of reservation can start the prep work once the reservation window opens and makes sure things are allocated accordingly before time X. 

We really appreciate your inputs.  We are currently mapping out the technical specs capturing the implementation models.  We'd be happy to share our approach as we refine them to ensure we take into account other considerations we fail to take into account for the community at large.