Details
-
Improvement
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
None
Description
Currently Thor can be configured to run multiple thorslaves per node with slavesPerNode.
That means none of the slaves can share any resources.
Having each thorslave process run multiple virtual slaves within the same process, would allow sharing of, e.g.:
1) Lookupjoin RHS between virtual slaves.
2) Keyed Join node caching
3) Shared single roxiemem memory pool.
4) Share AST of parse
5) .. other things I haven't thought of.
Attachments
Issue Links
- is duplicated by
-
HPCC-2623 Thor multislave optimizations
- Resolved
1.
|
Multiple slaves TOPN should possibly run on a single thread |
|
Resolved | Unassigned |
2.
|
Add method to get raw cluster group |
|
Resolved | Jake Smith |
3.
|
Enable ability to create/use multiple MP servers |
|
Resolved | Jake Smith |
4.
|
Use separate IRowManager's per channel to avoid contention |
|
Resolved | Jake Smith |
5.
|
Introduce a row manager that will be shared amongst slave channels |
|
Resolved | Jake Smith |
6.
|
Virtual Slave Thor supporting documentation |
|
Resolved | Jim DeFabia |