Shameless Statements about Replication

1
Shameless Statements about Replication
Rachid Guerraoui
School of Computer and Communication Sciences,
EPFL
Joint ruminations with Eli Gafni (UCLA-MSR)
2
Replication is all over the place

• Replicated databases

• Group communication

• Reliable middleware

• Storage systems

• Non-blocking data structures

3
Deconstructing replication
Highlight the important principles
(results/algorithms)
4
A perspective on replication
5
A grain of salt
Here are my principles. If you dont like them,
I have others
Groucho Marx
6
Deconstructing replication

• For now, lets
• (1) ignore performance

• and focus on
• (2) strong and general replication of

• (3) an object shared by 2 processes

7
Replication
opA
opA
P1
O
P2
opB
opB
8
(No Transcript)
9
(No Transcript)
10
Replication
P1
opB
opA
opA
O
Fair agreement on the order
P2
O
opB
opA
opB
11
Replication
P1
opB
opA
opA
O
Shared memory
Consensus
P2
O
opB
opB
opA
12
Statement

• (1) Behind every replication lie a
• consensus and a shared memory

Consensus
13
Consensus is impossible FLP

Asynchronous shared memory system
p1
p2
14
Synchronous consensus is possible
Synchronous system
Asynchronous system
15
ltgtSynchronous consensus is possible
ltgtSynchronous system
Asynchronous system
16
Consensus is almost possible
?-synchronous system
17
One person is missing and the whole world seems
depopulated Alphonse de
la Martine
p2
p1

• Consensus is possible iff ?-synchrony

18
Bottom line

• (1) Behind every replication lie a
• consensus and a shared memory

• (2) Behind every consensus lies some
• ?-synchrony

?-synchrony
19
(No Transcript)
20
Consensus (primary)
P1
V1
V1
Shared memory
P2
V2
V1
21
Consensus (primary)
P1
crash
V1
V1
Shared memory
P2
V2
22
Consensus (primary-backup)
P1
V1
V1
V1
Shared memory
P2
V2
V1
23
Consensus (primary-backup)
P1
crash
V1
V1
Shared memory
P2
V2
V2
24
Consensus (primary-backup)
P1
V1
V1
V1
Shared memory
P2
V2
V2
V2
25
Consensus (2PC)
P1
commit
V1
V1
V1
Shared memory
P2
V2
V1
26
Consensus (2PC)
P1
V2
V1
Shared memory
P2
V2
commit
V2
V2
27
Consensus (2PC)
P1
V1
abort
V1
Shared memory
P2
V2
abort
V2
28
Towards indulgent consensus

Asynchronous system

• Indulgence tolerates arbitrarily long periods of
  asynchrony, i.e., tolerates any prefix

29
Indulgence
30
Indulgence
 He that is without sin among you, let him cast
the first stone at her   John 83-11
31
Indulgence
Always preserves safety
Ensures liveness whenever possible

•  While there is life there is hope  Cicero

32
Indulgent consensus (3PC)
P1
V1
commit/abort
commit/abort
P2
V2
33
Indulgent consensus (3PC)

• The processes dynamically exclude one suffix of a
  run, using a system oracle

A failure detector
34
Weakest failure detector

• The weakest failure detector question translates
  into the smallest suffix set to be excluded

• The weakest failure detector for consensus - ?

35
Weakest failure detector
p1
p2
36
Shared memory assumption

• Helps better understand consensus results (FLP,
  FD, 2PC, 3PC)

Needed anyway for replication (and indulgent
consensus)
37
Bottom line

• (1) Behind every replication lie a
• consensus and a shared memory

• (2) Behind every consensus lies some
• ?-synchrony

shared memory
38
ABD (Snapshot)
write
P1
V1
Quorum
P2
V2
read
39
The many faces of quorums
Byzantine quorums
Probabilistic quorums
Failure detector quorums
Refined quorums
40
Bottom line

• (1) Behind every replication lie a
• consensus and a shared memory

• (2) Behind every consensus lies some
• ?-synchrony

• (3) Behind every shared memory lies
• a quorum

41
Deconstructing replication

• (1) ignoring performance

• And focusing on
• (2) strong and general replication of
• (3) one object shared by 2 processes

42
The engineer
Much ado about nothing?

• In real systems, we do care about performance and
  we are happy with weak replication

43
What about performance?
Lets move now to a message passing system with
communication delays/rounds
44
What about performance?
Synchronous system with few failures
Asynchronous system

• Plan for the worst and hope for the best

45
What about performance?

• How many synchronous rounds does consensus need?
  

At least t1
A shared memory system of n processes with 1
failure can simulate x rounds of a synchronous
system with x failures
46
The inherent price of indulgence

• How many synchronous rounds does an indulgent
  consensus need to decide with f failures?

At least f2
A shared memory system of n processes with 1
failure can simulate x1 synchronous rounds of an
indulgent consensus algorithm with x failures
47
The inherent price of indulgence

• For how long does a system need to be synchronous
  for indulgent consensus to terminate?

No clue
48
The inherent price of indulgence

• How many servers need to be correct in order for
  indulgent consensus to decide in x synchronous
  rounds?

Refined quorums
49
More about performance

• Disk accesses?
• Throughput?

50
What about weak replication?

• Is consensus necessary for weak replication?

If replicas would never need to agree on any
state, they would not be called replicas
51
What is weak replication?
The answer, my friend, is blowin' in the wind
52
What about weak replication?
Does ad-hoc replication need consensus?
Say we know the semantics of an object, e.g., a
queue? (weaker than consensus)
We need consensus among 2 processes
53
What about weak replication?

• Does eventual replication need consensus?

• It does eventually..

• Does probabilistic replication need consensus?

• It does need randomized consensus..

54
What if

• We give up safety and let some of the replicas
  disagree?

• We might need set-agreement

• We give up liveness and ensure termination only
  if k processes are concurrent?

• We might need set-agreement

55
The future of replication

• What form of quorum (shared memory) does a
  set-agreement actually need?

• For how long does a system need to be synchronous
  for indulgent set-agreement to terminate?

56
The one slide to remember

• (1) Behind every replication lie
• agreement and shared memory

• (2) Behind every agreement lies
• ?-synchrony

• (3) Behind every shared memory lies
• a quorum

57
Or at least this one
58
What about more processes?
n-process (f-1)-resilient system
f-process wait-free system