Causal Logging : Manetho

1
Causal Logging Manetho

• Rohit C Fernandes
• 10/25/01

2
Manetho System Model

• Non determinististic events
• Message Receive
• Internal event(Kernel call)
• Creation of a new process
• Output Commit
• Stable Storage Volatile Memory

3
Manetho properties

• Tolerate any number of simultaneous failures
• Low failure-free overhead
• Only failed processes roll back

4
Example Manetho Execution
5
Causal Logging Intuition

• Piggyback determinant of non-deterministic event
  on outgoing messages
• Determinant?
• Piggyback Antecedence Graphs

6
Antecedence Graph

• Directed acyclic graph
• Nodes State Intervals
• Edges Happened before(immediate)

7
Antecedence Graph
8
Receive Node

• Two incoming edges
• Fields
• Receiver ID
• Sender ID
• Index of created state interval
• Unique identifier of message

9
Internal Event Node

• One incoming edge
• Fields
• Type of event
• Replay information

10
Failure Free Operation

• Each process maintains
• AG of its current interval
• Log that contains data and ID of each message
  sent
• Message Send Piggyback AG of current state
  interval

11
Optimization

• Need not send complete AG
• Incremental piggybacking
• AG(?i1p) is a proper subgraph of AG(?ip)
• Process q communicates to p max j such that ?jp
  is in qs AG
• P sends AG (?ip ) - AG (?jp )

12
Information on Stable Storage

• Checkpoints
• AG (asynchronously) Need not piggyback part of
  AG which is in disk
• Output commit Save AG to disk

13
Incarnation Numbers

• Each process starts a new incarnation after
  recovery
• Integer stored in stable storage
• Tagged on outgoing messages
• Messages from old incarnations discarded

14
Recovery Protocol

• Recover(p,c,INCNUM,S)
• Step 1
• INCNUM ? INCNUM1 save INCNUM
• INCVECp? INCNUM
• G? AG(?pc) // stable storage

15
Recovery Protocol

• Step 2
• For all q ? S, q?p
• (INQ,AGQ)?remote call at qGET_AG(p)
• G?G?AGQ
• INCVECq?INQ
• For all q ? S, q?p
• Remote call at q SEND_INC(p,INCVEC)

16
Recovery Protocol

• Step 3
• m?max j such that ?pj ? G
• Recover upto ?pm
• Dont send out application messages but log them
• For receive, request message from senders log
• Replay internal event

17
Recovery Example
18
Available Antecedence Graphs
19
Application Characteristics
20
Performance Overhead
21
(No Transcript)
22
Coordinated vs. Uncoordinated
23
(No Transcript)