IMSuite: IIT Madras Benchmark Suite for Simulating
Distributed Algorithms

IMSuite Characterization

IMSuite: IIT Madras benchmark suite for simulating distributed algorithms, implements some of the classical distributed algorithms. These classical algorithms cover important characteristics of distributed systems such as communication (broadcast, unicast, or multicast), timing (synchronous, asynchronous or partially synchronous) and failure. These algorithms have been implemented as kernel benchmarks.

Details regarding IMSuite benchmarks can be studied here.
These kernel benchmarks can be downloaded from here.

IMSuite Contributions

IMSuite aims at delivering following set of resources and functionalities to the parallel and distributed systems community:

1. Implementation of twelve classical distributed algorithms in three task parallel languages X10, HJ and OpenMP. The respective reference serial implementations are also available.
2. Multiple variations (Varying Synchronization primitives and Forms of prallelization) of the kernels.
3. Algorithm specific input generator.
4. Output validator to verify the resulting output from the benchmarks.
5. Characterization of different aspects of distributed applications (or the simulations thereof):
   1. Behavior with increased problem size.
   2. Impact of variation in input configuration.
   3. Distribution of remote communication requests.
   4. Amount of synchronization operations.
   5. Scalability of the kernels with increase in the number of compute threads and number of clusters.
6. Characteristics specific to parallel systems.
   1. Number of task creation operations.
   2. Number of task termination operation.
   3. Number of task join operations.
   4. Number of atomic operations.