Feature #11601
open
- Target version deleted (
v9.0.3)
The most obvious problem is that crush/ uses raw pointers which seems to preclude putting the OSDMaps in memory shared across OSDs. Hmm
Do you mean we need to share a memory region between different osd processes on the same host?
- Project changed from Ceph to RADOS
- Subject changed from Share cached osdmaps across osd daemons to osd: share cached osdmaps across osd daemons
Attached the file CephScaleTestMarch2015.pdf
Do we have any plan for this guys?
A vague possibility that the future seastar-based OSD may run each logical disk OSD inside a single process, which could share maps — but nothing concrete yet.
Also available in: Atom
PDF