Mumbai University > Computer Engineering > Sem 8 > parallel and distributed systems

Marks: 10M

To reduce access time of data caching is used. The effect of replication and caching increases complexity and overhead of consistency management.

Different Consistency models offer different degrees of consistency.

Depending on the order of operations allowed and how much inconsistency can be tolerated, so consistency models range from strong to weak.

Data centric model:

In this model it is guarantee that the results of read and write operations which is performed on data can be replicated to various stores located nearby immediately.

Client centric model:

These consistency model do not handle simultaneous updates.

But, to maintain a consistent view for the individual client process to access different replicas from different locations has been carried out.

Data centric consistency models explanation

Strict consistency:

It is the strongest data centric consistency model as it requires that a write on a data be immediately available to all replicas.

This model states that “Any read on data item x returns a value corresponding to the result of the most recent write on x.

In a distributed system, it is very difficult to maintain a global time ordering, along with the availability of data which is processed concurrently at different locations, which causes delays for transferring of data from one location to another.

Thus Strict consistency model is impossible to achieve.

Eg P1 & P2 are two processes. P1 performs a write operation on its data item x and modifies its value to a.

Update is propogated to all other replicas

Suppose if P2 now reads and finds its value to be NIL due to the propogation delay.

So the result of this value read by P2 is not strictly consistent.

But as the law of Strictly consistent model says that results should be immediately propogated to all the replicas as shown in figure below

Client centric model explanation

Monotonic Reads:

It states that “If a process P reads the value of a data item t, any successive read operation on x by that process at later time will always return that same or a recent value.

For eg Each time one connects to the email server (may be different replicas),the email server guarantees that all the time will always return that same or recent value.

Here, in fig (a) operations WS(x1) at L1 before the second read operation. Thus WS(x1,x2) at L2 would see the earlier update.