Editing Scientific Data Manager (SDM) for Irregular Applications

Title: Scientific Data Manager (SDM) for Irregular Applications

Abstract:
The Scientific Data Manager (SDM) is a software system designed to manage large, scientific data sets generated by irregular applications. It uses a combination of parallel file I/O and database support for high-performance scientific data management. SDM provides a high-level API to the user and internally uses a parallel file system to store real data and a database to store application-related metadata. In this paper, we describe the design and implementation of SDM for irregular applications, focusing on its user interface and performance optimization techniques. We present performance results for two irregular applications, FUN3D and a Rayleigh-Taylor instability code, on the SGI Origin 2000 at Argonne National Laboratory.

Main Research Question:
How can we design and implement a software system that efficiently manages large, scientific data sets generated by irregular applications, while providing a convenient high-level API for users?

Methodology:
To answer this question, we followed these steps:

1. Identified the main objectives: We aimed to achieve high-performance I/O, provide a convenient high-level API, and optimize the execution cost of irregular applications.
2. Chose a parallel file-IO system and MPI-IO: We decided to use a parallel file-IO system to store real data and MPI-IO to access this data. MPI-IO is specifically designed to enable optimizations critical for high-performance applications.
3. Designed the SDM user interface: We designed a user-friendly interface that hides the details of I/O operations, making it easy for users to access data.
4. Implemented performance optimization techniques: We implemented techniques such as noncontiguous requests and collective I/O in MPI-IO to optimize data access and distribution.
5. Used a history file concept: We used the concept of a history file to optimize the cost of index distribution using metadata stored in the database.

Results:
We presented performance results for two irregular applications, FUN3D and a Rayleigh-Taylor instability code, on the SGI Origin 2000 at Argonne National Laboratory. The results showed that SDM can efficiently handle data reading and writing in an irregular mesh and distribute index values, achieving high-performance I/O.

Implications:
The implementation of SDM for irregular applications has several implications:

1. Efficient management of large, scientific data sets: SDM can handle data sets spanning hundreds or thousands of files, making it suitable for large-scale scientific applications.
2. High-performance I/O: By using a parallel file-IO system and MPI-IO, SDM can provide high-performance I/O for irregular applications.
3. Convenient high-level API: SDM's user-friendly interface hides the details of I/O operations, making it easy for users to access data.
4. Optimization of execution cost: SDM's performance optimization techniques, such as noncontiguous requests and collective I/O, can improve the execution cost of irregular applications.

Related Work:
We discussed related work in the field of scientific data management and irregular applications, focusing on previous research and existing software systems.

Conclusion:
In conclusion, SDM is a powerful tool for managing large, scientific data sets generated by irregular applications. Its high-performance I/O, convenient high-level API, and performance optimization techniques make it an ideal choice for large-scale scientific applications.

Link to Article: https://arxiv.org/abs/0102016v1
Authors: 
arXiv ID: 0102016v1

[[Category:Computer Science]]
[[Category:Data]]
[[Category:Applications]]
[[Category:Irregular]]
[[Category:Performance]]
[[Category:Sdm]]