Browse Prior Art Database

High Performance Computing based on Asynchronous Messaging for Novel Drug Discovery, Design and Development.

IP.com Disclosure Number: IPCOM000010690D
Original Publication Date: 2003-Jan-08
Included in the Prior Art Database: 2003-Jan-08
Document File: 4 page(s) / 137K

Publishing Venue

IBM

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 50% of the total text.

Page 1 of 4

  High Performance Computing based on Asynchronous Messaging for Novel Drug Discovery, Design and Development.

   Typical Inter-process communication (IPC) mechanisms between a Research Application Server and HPC (High Performance Computing) Servers are based on Synchronous RPC (Remote Procedure Call) paradigm. Consequently, some HPC servers are saturated while others are under-utilized. Furthermore, Pipeline operation is interrupted when a HPC node fails even if other servers are available. As a result, the expensive resources for compute intensive or data intensive processing are not well utilized for high performance and high availability for Pipeline management. Also, large volume of data to be processed by an HPC server is retrieved by the Research Application Server and sent as RPC input data. As a result, the network traffic between the Research Application Servers and HPC servers can become a bottle-neck.

The invention is a new inteprocess communication (IPC) mechanism for integration of various applications used for the pipeline management in drug discovery, design and development, leveraging the asynchronous messaging paradigm, or IBM MQSeriess, such that the workload among the HPC servers is dynamically managed and data to be processed can be retrieved by an HPC server itself. Specifically this design provides for workloads among the HPC servers are dynamically managed; availability of HPC servers are maximized; data traffic between Research Application Server and HPC servers are minimized; and business objects running on the Research Application Server and algorithms requiring compute intensive and/or data intensive processing can be independent of operating platforms. Figure 1 illustrates the problem.

Typical IPC Mechanism

Data-1 Data-2 Data-3 Data-4

Algorithm

A

Algorithm

B

 Research Application Server

Algorithm

C

Algorithm

D

Data-1, Data-2,

Data-3, Data-4

HPC Servers

OKB 03172K2

Figure 1

As shown in Figure 1, the same network medium needs to be shared for the flow of data-1, data-2, data-3, and data-4. Also, when the HPC node running Algorithm-A is down, the pipeline involving the algorithm must be interrupted.

1

[This page contains 6 pictures or other non-text objects]

Page 2 of 4

Figure 2 and Figure 3 illustrate the interprocess mechanism based on asynchronous messaging between Research Application Server and HPC servers to maximize performance and availability of pipeline management for drug discovery, design and development.

New IPC Mechanism

Result Queues

Algorithms and Data are dynamically loaded onto HPC servers based on the WorkObjects in the queue

Figure 2

Research Applications put a WorkObject in the job queue, specifying the algorithm name and the location of the data to be processed. The Queue manager will pick up the WorkObjec...