Reliable Server Pooling Architecture

1
Reliable Server PoolingArchitecture

• Michael Tüxen
• tuexen_at_fh-muenster.de

2
The Motivation

• Build systems without a single point of failure
  by using servers not being highly reliable.
• SCTP addresses network failures.
• RSerPool addresses host failures (as much as
  possible).

3
Server Pools

• Use a set of servers instead of one.
• Identify them by a Pool Handle (PH).
• Provide a mechanism to resolve PH to transport
  addresses.
• State sharing between the servers is out of scope
  of RSerPool

4
Entities in the Architecture

• Pool Elements (PEs)
• the servers in a server pool.
• Pool Users (PUs)
• the clients.
• Name Servers (NSs)
• the entities performing the name resolution.

5
Communication relations
PU
PU
6
Protocols

• Aggregate Server Access Protocol (ASAP)
• PE lt-gt NS
• PU lt-gt NS
• PU lt-gt PE (Enhanced Mode)
• Endpoint Name Resolution Protocol (ENRP)
• NS lt-gt NS

7
What are the protocols used for?

• ASAP
• Registration of PEs.
• Re-registration of PEs.
• Supervision of PEs.
• Name resolution.
• Failover support.
• ENRP
• Complete state sharing between the NSs.

8
General Considerations

• Lightweight Client have to be supported.
• Two entities are in the transport address
  selection involved.
• There are other applications for RSerPool
  distributed computing, support for transport
  layer mobility, ...

9
Failover Support

• Business Cards for PUs being PEs.
• Last Will for preferred PEs to fail over.
• State Cookie for state synchronization.
• Only available in Advanced Mode.