Synchronous Atomic Broadcast Implementation
Report by Jacob Schrum
for CS386C Spring 2009
This project was implemented as a PHP webpage in order to make
it easily accessible to anyone interested in understanding how
synchronous atomic broadcast works. The simulator can be accessed
and used by anyone at my webpage:
http://www.cs.utexas.edu/~schrum2/cs386c/sab.php
The code models synchronous atomic broadcast between processors
in a network with redundant channels. The simulator also allows
the user to set up several types of faults in advance, in order
to test the correctness of the system. The simulator uses the
model proposed by Flaviu Cristian, but a few simplifying
assumptions are also made. The properties of the system are
as follows:
Clock drift ignored: Clock drift of processors is not
modelled. Given a starting time, processors will reliably maintain
time relative to their starting time
Clock synchronization assumed: Although there is no
clock drift, different clocks are allowed to have different values.
These starting values all have to be within epsilon time units of
each other. Because there is no drift, this means the clocks will
remained synchronized within epsilon.
Initialize at 0: The starting global time of the simulation
is always assumed, without loss of generality, to be 0. The local
clocks are also synchronized within epsilon of the global time.
Therefore all clocks start within times in [0,epsilon].
Sends occur simultaneously: When a processor sends
a message, either by initiating it or by forwarding, it will
typically need to send that message on more than one channel.
Although sends to higher channels technically occur after sends
to lower channels, the simulation processes all of these commands
at the same time, meaning they all occur at the same time step
of the simulation. However, processor crashes during sends can still
occur between broadcasts.
Random transmission times: The timeout on channel
broadcasts is delta. Whenever a processor executes a command
to broadcast on a particular channel, a random integer in [0,delta]
is chosen for each receiving processor as the transmission time.
Maximum faults defined for whole simulation: The
maximum number of allowable faults f per broadcast is in
this simulation also used to define the total number of
faults during the simulation period. This was done in part to
simplify the user interface, and in part because some
faults are designed to affect several broadcasts.
Channel omissions occur across all messages on channel:
One of the allowable fautl types is a channel omission. This
will affect all messages currently on the channel. Such an omission
will only ever be a single fault to a particular broadcast,
but will impact every individual broadcast curently making use
of the channel.
Certain faults are scheduled in advance: The
way that the user will specify most faults will be to
designate a global time after which they will occur. For
example, crashes while forwarding and in-adapter omissions
are scheduled in advance, but they will only occur if
conditions allow. For example, if a processor's in-adapter is
scheduled to omit the first message after a certain time t,
but doesn't receive any messages after that time, then
the fault will never occur.
For more information regarding the source code and the
user interface, please refer to the
user manual.