cugraph
cugraph copied to clipboard
[EPIC] Optimize renumbering
The current renumbering performs cudf/dask_cudf merges for unrenumbering and re-renumbering. Single column renumbering can be optimized using local unrenumbering to make it faster instead of merge operation
From @seunghwak - to help address this: https://github.com/rapidsai/cugraph/blob/branch-21.12/cpp/include/cugraph/graph_functions.hpp#L199 https://github.com/rapidsai/cugraph/blob/branch-21.12/cpp/include/cugraph/graph_functions.hpp#L229 https://github.com/rapidsai/cugraph/blob/branch-21.12/cpp/include/cugraph/graph_functions.hpp#L298 We should unrenumber using these C++ functions instead of relying on dask-cudf.
For integer vertex columns we can address this in the new C API.
This issue has been labeled inactive-30d
due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d
if there is no activity in the next 60 days.
I believe that all of the plumbing exists in C++ to handle this properly for int32 and int64 types. IMO, there are two options here:
- The C API already does this for int32 and int64 types, we can python implementations to the C API and get this for free
- We can modify the NumberMap.py implementation to call the C++ mechanisms which would get us all of the algorithms doing this now, at the cost of some throw away code
I bumped this out to 22.04, but 22.06 might be a better target. Alternatively, we could close this in favor of migrating everything to the C API.
This issue has been labeled inactive-30d
due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d
if there is no activity in the next 60 days.
This issue has been labeled inactive-30d
due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d
if there is no activity in the next 60 days.
This issue has been labeled inactive-90d
due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.