Most of the mathematical models in finance are highly resource intensive, when it comes to processing. Therefore, it is encouraging to go for a High Performance Computing resolution. As a result, We, Damitha, Nilendra, Thilina and myself (Waruna) came up with a system which consists of a computational cluster, database cluster and an Excel plug-in to speed-up the financial calculations as well as to add more user convenience by providing a user interface as an Excel plug-in. Hence your results are only a click away.
The system is developed in such a way that other financial models can be integrated to the system dynamically by the user (in this case the developers). In the latest release of HPC4Finance has inbuilt support for three (3) short rate models, namely Geometric Brownian Motion (GBM), Vasicek Model and Cox-Ingersoll-Ross model (CIR).
MPICH2 (1.0.8) cluster which is an implementation of MPI 2.0 specification, is used to implement the computational cluster. The cluster is built on UNIX environment. But it is available for other platforms as well (i.e. Windows even though its not tested enough). The two major problems, we faced with the MPICH2 cluster are load balancing and the fault tolerance. A simple resolution is introduced to improve load balancing by using a round robin algorithm to resolve the jobs (tasks) and their respective execution nodes. Say, that 2 jobs where both are supposed to run in 10 machines (nodes), are submitted to the cluster where there are 20 machines. Then, in default case, both the jobs will run in the same set of nodes and there are 10 nodes idling. This is a huge waste of resources. Therefore, as our resolution, the jobs are explicitly submitted to the next (available) set of nodes in a round robin manner. Unfortunately, we could not come up with an straight forward mechanism to fault tolerance and fail over management which should be a functionality of the cluster itself. But we provide a logging system which enables to track the fault and restore or restart the job.
MySQL cluster is used to implement the database cluster. The data is automatically replicated through the cluster. The cluster support data node fail-over, but not SQL (API) node fail-over.
Excel plug-in is the most important component which adds a great deal of value to the system. It enables financial professionals who are not keen in programming and computer clusters to run their financial models by using just a few mouse clicks. The results are drawn on a graph and presented to the user with the other relevant information. The plug-in enables users to import and update financial data through a web service or form a .csv file.
The user is also facilitated with a ribbon, to change and the set the preferences and settings of the plug-in.
The user also equiped with ability to stop (kill) jobs any time which enables users to terminate jobs whose parameters are entered incorrectly.
The system (HPC4Finance) is just an insight to the possibilities of High Performance Computing to the field of Finance with great deal of extendibility and user convenience.
Reference
MPICH2 (1.0.8) cluster which is an implementation of MPI 2.0 specification, is used to implement the computational cluster. The cluster is built on UNIX environment. But it is available for other platforms as well (i.e. Windows even though its not tested enough). The two major problems, we faced with the MPICH2 cluster are load balancing and the fault tolerance. A simple resolution is introduced to improve load balancing by using a round robin algorithm to resolve the jobs (tasks) and their respective execution nodes. Say, that 2 jobs where both are supposed to run in 10 machines (nodes), are submitted to the cluster where there are 20 machines. Then, in default case, both the jobs will run in the same set of nodes and there are 10 nodes idling. This is a huge waste of resources. Therefore, as our resolution, the jobs are explicitly submitted to the next (available) set of nodes in a round robin manner. Unfortunately, we could not come up with an straight forward mechanism to fault tolerance and fail over management which should be a functionality of the cluster itself. But we provide a logging system which enables to track the fault and restore or restart the job.
MySQL cluster is used to implement the database cluster. The data is automatically replicated through the cluster. The cluster support data node fail-over, but not SQL (API) node fail-over.
Excel plug-in is the most important component which adds a great deal of value to the system. It enables financial professionals who are not keen in programming and computer clusters to run their financial models by using just a few mouse clicks. The results are drawn on a graph and presented to the user with the other relevant information. The plug-in enables users to import and update financial data through a web service or form a .csv file.
The user is also facilitated with a ribbon, to change and the set the preferences and settings of the plug-in.The user also equiped with ability to stop (kill) jobs any time which enables users to terminate jobs whose parameters are entered incorrectly.
The system (HPC4Finance) is just an insight to the possibilities of High Performance Computing to the field of Finance with great deal of extendibility and user convenience.
Reference
T.P. Dampahala, H.D.D.D. Premadasa, P.W.W. Ranasinghe, J.N.P. Weerasinghe, K.A.D.N.K. Wimalawarne, "Efficient High Performance Computing Framework for Short Rate Models," ams, pp.608-613, 2009 Third Asia International Conference on Modelling & Simulation, 2009
Can be found at - Efficient High Performance Computing Framework for Short Rate Models - http://doi.ieeecomputersociety.org/10.1109/AMS.2009.27
List of published papers in the 3rd Asia International Conference on Modeling & Simulation
Can be found at - Efficient High Performance Computing Framework for Short Rate Models - http://doi.ieeecomputersociety.org/10.1109/AMS.2009.27
List of published papers in the 3rd Asia International Conference on Modeling & Simulation
No comments:
Post a Comment