There is a reference Newtonian real-time t
(UTC).
No process has direct access to this reference time, used only for specification purposes.
Processes are assumed to be equipped with synchronized clocks.
This requires processes to periodically synchronize their local clocks with an external and trusted source of the time (e.g. NTP servers). Each synchronization cycle aligns the process local clock with the external source of time, making it a fairly accurate source of real time. The periodic (re)synchronization aims to correct the drift of local clocks, which tend to pace slightly faster or slower than the real time.
To avoid an excessive level detail in the parameters and guarantees of
synchronized clocks, we adopt a single system parameter PRECISION
to
encapsulate the potential inaccuracy of the synchronization mechanisms,
and drifts of local clocks from real time.
There exists a system parameter PRECISION
, such that
for any two processes p
and q
, with local clocks C_p
and C_q
,
that read their local clocks at the same real-time t
, we have:
p
and q
are equipped with synchronized clocks, then |C_p(t) - C_q(t)| < PRECISION
PRECISION
thus bounds the difference on the times simultaneously read by processes
from their local clocks, so that their clocks can be considered synchronized.
The first draft of this specification included a second clock-related parameter, ACCURACY
,
that relates the values read by processes from their synchronized clocks with real time:
p
is a process is equipped with a synchronized clock, then at real time
t
it reads from its clock time C_p(t)
with |C_p(t) - t| < ACCURACY
The adoption of ACCURACY
as the upper bound on the difference between clock
readings and real time, however, renders the PRECISION
parameter redundant.
In fact, if we assume that clocks readings are at most ACCURACY
from real
time, we would therefore be assuming that they cannot be more than 2 * ACCURACY
apart from each other, thus establishing a worst-case upper bound for PRECISION
.
The approach we take is to assume that processes clocks are periodically
synchronized with an external source of time, thus improving their accuracy.
This allows us to adopt a relaxed version of the above ACCURACY
definition:
t
there is at least one correct process p
which clock marks
C_p(t)
with |C_p(t) - t| < ACCURACY
Then, through [PBTS-CLOCK-PRECISION] we can extend this relation of clock times
with real time to every correct process, which will have a clock with accuracy
bound by ACCURACY + PRECISION
.
But, for the sake of simpler specification we can assume that the PRECISION
,
which is a worst-case parameter that applies to all correct processes,
includes the best ACCURACY
achieved by any of them.
The assumption that processes have access to synchronized clocks ensures that proposal times assigned by correct processes have a bounded relation with the real time. It is not enough, however, to identify (and reject) proposal times proposed by Byzantine processes.
To properly evaluate whether the time assigned to a proposal is consistent with the real time, we need some information regarding the time it takes for a message carrying a proposal to reach all its (correct) destinations. More precisely, the maximum delay for delivering a proposal to its destinations allows defining a lower bound, a minimum time that a correct process assigns to proposal. While minimum delay for delivering a proposal to a destination allows defining an upper bound, the maximum time assigned to a proposal.
There exists a system parameter MSGDELAY
for end-to-end delays of messages carrying proposals,
such for any two correct processes p
and q
, and any real time t
:
p
sends a message m
carrying a proposal at time ts
,
then if q
receives the message and learns the proposal,
q
does that at time t
such that ts <= t <= ts + MSGDELAY
.While we don't want to impose particular restrictions regarding the format of m
,
we need to assume that their size is upper bounded.
In practice, using messages with a fixed-size to carry proposals allows
for a more accurate estimation of MSGDELAY
, and therefore is advised.
In this section we define the properties of Tendermint consensus (cf. the arXiv paper) in this new system model.
A proposer proposes a consensus value v
with an associated proposal time v.time
.
[Agreement] No two correct processes decide on different values v
. (This implies that no two correct processes decide on different proposal times v.time
.)
[Validity] If a correct process decides on value v
,
then v
satisfies a predefined valid
predicate.
[Time-Validity] If a correct process decides on value v
,
then the associated proposal time v.time
satisfies a predefined timely
predicate.
Both [Validity] and [Time-Validity] must be observed even if up to
2f
validators are faulty.
The timely
predicate is evaluated when a process receives a proposal.
Let now_p
be time a process p
reads from its local clock when p
receives a proposal.
Let v
be the proposed value and v.time
the proposal time.
The proposal is considered timely
by p
if:
now_p >= v.time - PRECISION
andnow_p <= v.time + MSGDELAY + PRECISION
We denote by POL(v,r)
a Proof-of-Lock of value v
at the round r
of consensus.
POL(v,r)
consists of a set of PREVOTE
messages of round r
for the value v
from processes whose cumulative voting power is at least 2f + 1
.
If
POL(v,r*)
for height h
, andr*
is the lowest-numbered round r
of height h
for which there is a valid POL(v,r)
, andPOL(v,r*)
contains a PREVOTE
message from at least one correct process,Then, where p
is a such correct process:
p
received a PROPOSE
message of round r*
and height h
, andPROPOSE
message contained a proposal for value v
with proposal time v.time
, andp
considered the proposal timely
.The round r*
above defined will be, in most cases,
the round in which v
was originally proposed, and when v.time
was assigned,
using a PROPOSE
message with POLRound = -1
.
In any case, at least one correct process must consider the proposal timely
at round r*
to enable a valid POL(v,r*)
to be observed.
If
POL(v,r)
for height h
, andPOL(v,r)
contains a PREVOTE
message from at least one correct process,Then
POL(v,r*)
for height h
, with r* <= r
, andPOL(v,r*)
contains a PREVOTE
message from at least one correct process, andv
timely
at round r*
.The above relation derives from a recursion on the round number r
.
It is trivially observed when r = r*
, the base of the recursion,
when a timely POL(v,r*)
is obtained.
We need to ensure that, once a timely POL(v,r*)
is obtained,
it is possible to obtain a valid POL(v,r)
with r > r*
,
without the need of satisfying the timely
predicate (again) in round r
.
In fact, since rounds are started in order, it is not likely that
a proposal time v.time
, assigned at round r*
,
will still be considered timely
when the round r > r*
is in progress.
In other words, the algorithm should ensure that once a POL(v,r*)
attests
that the proposal for v
is timely
,
further valid POL(v,r)
with r > r*
can be obtained,
even though processes do not consider the proposal for v
timely
any longer.
This can be achieved if the proposer of round
r' > r*
proposesv
in aPROPOSE
message withPOLRound = r*
, and at least one correct processes is aware of aPOL(v,r*)
. From this point, if a validPOL(v,r')
is achieved, it can replace the adoptedPOL(v,r*)
.
The safety of the algorithm requires a timely proof-of-lock for a decided value, either directly evaluated by a correct process, or indirectly received through a derived proof-of-lock.
If
C
for height k
and round r
, andC
contains a PRECOMMIT
message from at least one correct processThen, where p
is one such correct process:
p
is correct, p
received a valid POL(v,r)
, andPOL(v,r)
contains a PREVOTE
message from at least one correct process, andPOL(v,r)
is derived from a timely POL(v,r*)
with r* <= r
, andPOL(v,r*)
contains a PREVOTE
message from at least one correct process, andv
timely
at round r*
.In terms of liveness, we need to ensure that a proposal broadcast by a correct process
will be considered timely
by any correct process that is ready to accept that proposal.
So, if:
p
of a round r
is correct,POL(v',r')
for any value v'
and any round r' < r
,p
proposes a valid value v
and sets v.time
to the time it reads from its local clock,Then let q
be a correct process that receives p
's proposal, we have:
q
receives p
's proposal after its clock reads v.time - PRECISION
, andq
is at or joins round r
while p
's proposal is being transmitted,
then q
receives p
's proposal before its clock reads v.time + MSGDELAY + PRECISION
Note that, before
GST
, we cannot ensure that every correct process receivesp
's proposals, nor that it does it while ready to accept a roundr
proposal.
A correct process q
as above defined must then consider p
's proposal timely
.
It will then broadcast a PREVOTE
message for v
at round r
,
thus enabling, from the Time-Validity point of view, v
to be eventually decided.
MSGDELAY
sThe liveness assumptions of PBTS are conditioned by a conservative and clever
choice of the timing parameters, specially of MSGDELAY
.
In fact, if the transmission delay for a message carrying a proposal is wrongly
estimated, correct processes may never consider a valid proposal as timely
.
To circumvent this liveness issue, which could result from a misconfiguration,
we assume that the MSGDELAY
parameter can be increased as rounds do not
succeed on deciding a value, possibly because no proposal is considered
timely
by enough processes.
The precise behavior for this workaround is under discussion.
Back to main document.