|Main||Roadmap and Innovation||Technology||For Users||For Resource Providers||Media|
Scenario 4: Accounting across Resource Providers
Leader: Alison Packer and John Gordon, STFC
|Scenario leader||STFC||Alison Packer|
To account for resource usage across multiple resource providers we need to define:
- the elements to be accounted for and how accounting data may be gathered. (This relates to the statement “Resource providers agree and follow the same rules for accounting the resource usage”.)
- how accounting data will be published
- what is required from this data (for VM users, Resource Providers, VO Managers, others?)
Draft Usage Record
To help define what we will account for the existing Compute Accounting Usage Record  could be extended. The following are a set of assumptions I have made related to this draft Usage Record:
- For each instantiated VM there is one record, this has implications for charging as there is one ‘Charge’ attribute per record it does not fit with a charging model where each of the elements of CPU, Memory, I/O etc. are charged for separately.
- The fields for GlobalUserName, VirtualOrganization, Projectname and Group relate to an existing method of authorizing grid users, if this authorization mechanism were to be used for cloud usage as well then these would provide a method of associating VM usage with a user and then querying usage in a portal could be based on user, VO etc.
- A VM might be in one of three states: started, stopped and suspended. I think we would expect to gather many VM records whilst a VM is running, so I propose using the existing UR “Status” property for this – it may be possible to extend the implementation of this property to include some VM-specific statuses if these existing ones are not deemed appropriate.
- We need a method of identifying where a specific running VM is – there are many ways to do this, I suggest that the RecordIdentity might contain a concatenation of some of the other fields in the record – e.g. TimeStamp SiteName MachineName ImageID - other suggestions welcome.
- We have job accounting and storage accounting systems and their related usage records already - we therefore account for jobs and storage using these existing records/methods.
The following table shows what might be used and some additional elements required for cloud accounting:
|Current UR Property||Current Definition||Extended for Cloud Accounting|
|RecordIdentity||Unique identifier of a record (String)||(keep the same)|
|GlobalUserName||Global identity of user (certificate DN) (String)||(keep the same)|
|VirtualOrganization||All three may be retrieved from FQAN - use if VOs part of authorization mechanism||(keep the same)|
|Charge||Total charge of job (Float)|
|Status||Completion status (String), aborted, completed, failed, held, queued, started, suspended.||For VM instance - completed, started and suspended could be used|
|WallDuration||WallClock time||Is this EndTime in the case of VM?|
|CPUDuration||CPU time consumed (Duration)||(keep the same)|
|EndTime||Completion time (Timestamp)||Set to NULL until Status = completed|
|StartTime||Start time of the job (Timestamp)||Must be set if Status = started|
|MachineName||Hostname of the LRMS (String)||VM Hostname|
|Network||Amount of network resource used by the job (+ integer)||Network resource used by the VM|
|Memory||Amount of physical memory used by the job (+ integer)||Memory associated with the VM|
|TimeDuration||Additional measure of time duration (Duration)||This could be used for suspended VM to record suspended-endtime|
|SiteName||GOCDB SiteName||Extend service types in GOCDB to include cloud resource providers|
|Proposed New Property||Definition|
|ImageID||Every image has a unique ID associated with it|
|StorageRecordIdentity||Link to associated storage record|
Feedback from Resource Providers
|VM||What is measured|
|Oxford/Eucalyptus (no longer running)||Number of VM instantiated - Type of VM - Amount of time each VM has been running - Number of Elastic Block Storage volumes created - Amount of time the EBS volumes have been owned - Times the EBS volumes have been mounted - Binding between EBS volumes and VMs - Number of OS images loaded in the S3 repository - Amount of time the OS images have been loaded into the S3 repository|
|OpenNebula||(Have reviewed UR)|