Multicast collaborative erasure encoding and distributed parity protection

1. A method of multicast collaborative erasure encoding of a chunk stored in a distributed object storage cluster, the method comprising:

multicasting a roll-call request to every storage server in a negotiating group for the chunk;

generating and multicasting roll-call inventory responses by every storage server in the negotiating group; and

collecting the roll-call inventory responses by every storage server in the negotiating group from other storage servers in the negotiating group to form a set of roll-call inventory responses;

processing the set of roll-call inventory responses by every storage server in the negotiating group, wherein the processing diagnoses a state with a whole replica of the chunk held by at least one storage server and a replication count for erasure-encoded slices that is less than a target number of erasure-encoded slices; and

assigning actions to the storage servers holding a whole replica of the chunk to generate erasure-encoded slices and replicate the erasure-encoded slices to storage servers in the negotiating group that are selected to hold the slices.

2. The method of claim 1 , wherein the roll-call request comprises:

an identifier for the negotiating group;

an identifier for the chunk; and

an identifier for the roll-call request.

3. The method of claim 2 , wherein the roll-call request further comprises:

identification of a specific erasure-encoding algorithm to be used.

4. The method of claim 2 , wherein a roll-call inventory response multicast by a storage server comprises:

the identifier for the roll-call request;

an identifier for the storage server;

whether a whole replica of the chunk is stored at the storage server; and

which, if any, erasure-encoded slices of the chunk are stored at the storage server.

5. The method of claim 1 , wherein a roll-call inventory response that is missing due to not being collected by a storage server is ignored without a retransmission request.

6. The method of claim 1 , wherein the processing begins after the roll-call inventory responses have been received from every storage server in the negotiating group or a predetermined time has elapsed.

7. The method of claim 1 , further comprising:

each storage server in the negotiating group determining a subset of the actions to be performed by itself based on the set of roll-call inventory responses; and

each storage server in the negotiating group performing the subset of the actions to be performed by itself.

8. The method of claim 7 , further comprising:

selection of a storage server in the negotiating group to multicast a new roll-call request for the chunk.

9. The method of claim 8 , wherein the selection of the storage server in the negotiating group to multicast the new roll-call request comprises:

each storage server in the negotiating group pseudo-randomly selecting a time to multicast a new roll-call request for the chunk, wherein the new roll-call request is not issued if pre-empted by receipt of the new roll-call request as multicast by another storage server.

10. A distributed object storage system that stores objects in chunks, the system comprising:

a plurality of storage servers communicatively interconnected by a network; and

negotiating groups for the chunks,

wherein a negotiating group for a chunk comprises a group of the storage servers that are assigned to store and provide access to the chunk, and

wherein a storage server in the negotiating group for the chunk includes executable code that performs steps including:

multicasting a roll-call request to other storage servers in the negotiating group for the chunk;

collecting roll-call inventory responses received from the other storage servers in the negotiating group for the chunk to form a set of roll-call inventory responses;

processing the set of roll-call inventory responses, wherein the processing diagnoses a state with a whole replica of the chunk held by at least one storage server and a replication count for erasure-encoded slices that is less than a target number of erasure-encoded slices; and

assigning an action to the at least one storage server holding a whole replica of the chunk to generate an erasure-encoded slice and replicate the erasure-encoded slice to a storage server in the negotiating group that is selected to hold the slices.

11. The system of claim 10 , wherein the executable code performs further steps including:

receiving the roll-call request from another storage server in the negotiating group; and

multicasting a roll-call inventory response to the negotiating group in response to the roll-call request.

12. The system of claim 11 , wherein the roll-call request comprises:

an identifier for the negotiating group;

an identifier for the chunk;

an identifier for the roll-call request; and

identification of a specific erasure-encoding algorithm to be used.

13. The system of claim 11 , wherein the roll-call inventory response comprises:

the identifier for the roll-call request;

an identifier for the storage server;

whether a whole replica of the chunk is stored at the storage server; and

which, if any, erasure-encoded slices of the chunk are stored at the storage server.

14. The system of claim 10 , wherein the processing begins after the roll-call inventory responses have been collected from every storage server in the negotiating group or a predetermined time has elapsed.

15. The system of claim 10 , wherein the executable code performs further steps including:

determining a subset of the actions to be performed by the storage server based on the set of roll-call inventory responses; and

performing the subset of the actions.

16. The system of claim 15 , wherein one storage server in the negotiating group is selected to multicast a new roll-call request for the chunk.

17. The system of claim 16 , wherein the selection of the one storage server in the negotiating group to multicast the new roll-call request comprises:

each storage server in the negotiating group pseudo-randomly selecting a time to multicast a new roll-call request for the chunk, wherein the new roll-call request is not issued if pre-empted by receipt of the new roll-call request as multicast by another storage server.