Abstract:
As the e-commerce sites are being more secure and reliable in recent years
and the number of transactions is rising rapidly, parallelism can help us to
reduce response time and increase throughput for e-commerce transactions.
This paper will investigate parallelism in on-line transaction processing. It
aims to specify those aspects of e-commerce transactions that would profit
from parallel processing and analyze current parallel processing techniques
to determine those which can be used for e-commerce transactions. The
parallel processing techniques proposed in this paper can be easily applied to
B2C and B2B on-line transaction processing. Although some parallel
implementations of databases have been proposed, to the best of our
knowledge, parallel implementations of on-line transaction processing
specific to e-commerce are rarely existed.
Machine summary:
In this paper, we focus on using parallel processing techniques to improve the performance and throughput of e-commerce transaction processing systems.
Although some parallel implementations of OLAP transactions (Goi1 & Choudhary, l997/l999) and many parallel implementations of databases have been proposed, to the best of our knowledge, parallel implementations of OLTP transactions specific to e-commerce are rarely existed (Furtado, 2004; Dewitt & Gray, 1992; Raman, Han, & Narang, 2005; Wolf, Turek, Chen, & Ya, 1994).
This paper describes an e-commerce system design that is three-tiered architecture (the GUI tier, the business logic tier, and the database tier) system using different parallel processing techniques.
e. , search times were sustained when the number of processor nodes was increased in proportion to the amount of data in the database) were achieved for all searches using different one-criterion queries in various tests for Search1.
e. , t"' t”' it ) Hence, (View the image of this page) This explains why linear speedup was achieved (when the result set was small and the data for the search was large).
e. , t"' t”' it ) Hence, (View the image of this page) This explains why linear speedup was achieved (when the result set was small and the data for the search was large).
Hence, the run-time of a parallel order transaction can be computed using: (View the image of this page) For the test for small orders (shown in Figure 10), when the involving nodes increased, the communication cost increased and T ‹ -!!pd / n decreased.
(
, 
, 
)
