Speaker Andrew Tridgell
Time 2007-01-17 11:50
Conference LCA2007

CTDB - clustered TDB

Clustered Samba systems

Current - add mode nodes - slower

2 nodes should be faster then 1.

15 nodes needed to be faster then 1.

cifs = coherent

Behaviour exactly the same for multinode cluster as for single node system.

No shortcuts (unlike NFSv2 and v3).

Samba database sever:

  • mysql backend tdb

databases relatively slow

  • filesystem = database


  • very simple
  • very fast - 100K/second operations
  • byte range locking

Clustered databases too slow

tdb database running over clustered filesystem too slow.

  • robustness is high priority.

Robustness not a criteria for Samba databases 0 can look data.

If node lost, only data lost in data associated with node, and clients to node.

  • tdb stored in memory.
  • No data redundancy

Remote locking is evil, may case cascading locking affect

  • don’t hold lock when performing network operation.

  • send function/code to node that holds data

encryption slow


assume same version of Samba on all clusters processors.