I was going through the ScaleIO overview and the following questions popped up .
1) How the SDC build the metadata map ? Is the chunk info provided to SDC to build the map provided by the MDM or the SDS ? (I am assuming the info is provided by the MDM as it has all the metadata info)
2) Does the metadata map consist info regarding the primary Chunks only or the secondary as well ?
3) It says that the MDM uses a "lazy approach " to interact with the SDC and SDS. What exactly is this lazy approach ?
The reference to "lazy approach" can be found in the ScaleIO 1.30 training materials from EMC's Education site - Module 1, slide 20 from the older documents I have.
The term is referring to the fact that rather than proactively communicating/updating every SDC out there with an SDS chunk ownership change when it happens (could be a ton of clients out there), it will simply let the SDC contact the wrong SDS (worse-case) that doesn't own the chunk and reject the I/O - thus forcing the SDC to then contact the MDM to get an updated map of where the SDS-owner is and retry the I/O there for completion.
When an MDM cluster is created at that time Metadata map is created. As you go on adding the new SDS, the Metadata map is modified. A copy of metadata map is placed on each SDC so that SDC knows where the Chunk is to be sent and where is the chunk located.
echolaughmk
2 Intern
•
522 Posts
1546
0
Posted June 22nd, 2015 13:00
Hi Sanjeev,
The reference to "lazy approach" can be found in the ScaleIO 1.30 training materials from EMC's Education site - Module 1, slide 20 from the older documents I have.
The term is referring to the fact that rather than proactively communicating/updating every SDC out there with an SDS chunk ownership change when it happens (could be a ton of clients out there), it will simply let the SDC contact the wrong SDS (worse-case) that doesn't own the chunk and reject the I/O - thus forcing the SDC to then contact the MDM to get an updated map of where the SDS-owner is and retry the I/O there for completion.