|
Performance
Comparison of USB 2.0 vs. SATA II Analysis of Non-Corresponding Measurements Embry-Riddle Aeronautical University Project
Summary As the digital age matures, so do the habits and
savvy of everyday consumers enjoying its ever increasing technological
complexity. Music collections are transitioning from physical media to space
on a flash drive; home entertainment is recorded in high resolution format and
stored on hard disk drives. These escalating sizes in data are often impacted
by the bottleneck that occurs during data transfer. For example, a user may
want to transfer content from their home entertainment system to their
laptop, or archive it in an external hard drive. Often these mechanisms are
external attachments that require use of a universal/standardized means of
data transfer. For some time USB 2.0 was a dominant solution, and more
recently SATA II has moved in to become an alternative for high-bandwidth
data transfer. USB 2.0 claims to support data transfer rates up to 480Mbps,
and SATA II claims to support data transfer up to 3.0Gbps (using 10/8
bitstuffing – actual “throughput” is lower). However, these claims of maximum
performance may be lower in practice. The question is, with any statistically
significant measurement, does SATA II in practice exhibit better performance
as compared to USB 2.0 when transferring data across a cabled connection? The objective of this study is to use a formalized
experimental process to confirm or deny the hypothesis that SATA II will
perform better than USB 2.0 when transferring files across each respective
data bus. The determination will be based on a statistically sound analysis
of multiply-replicated measured values resulting from reproducible
experimental conditions. Moreover, the experiment aims to obtain a
quantitative value characterizing the performance difference between the
alternatives, or alternatively, explain the results if it is discovered that there
is not a statistically significant performance difference between SATA II and
USB 2.0. The system used to perform this experiment, consists
of a personal laptop with an internal SATA II hard drive, supporting both USB
2.0 an eSATA (external SATA) databus connection
capabilities. An external SATA-capable hard drive with both USB 2.0 and eSATA interfaces was connected to the host laptop via one
(and only one) of the possible connections during a given trial. Custom
software was developed to automate the transfer of the file from the laptop
to the external drive, recording the file transfer times into an external
log. The data collected in this log was imported into an Excel spreadsheet
for statistical analysis. Careful attention was given to reducing error by
making all reasonable attempts to recreate initial conditions and reduce
variation between experimental measurements. There were two factors in the experiment: the type
of bus being used and the size of the file transferred. These factors were
examined independently to eliminate the complexity of interactions between
the factors. This obviated the need for a two-factor Analysis of Variance
(ANOVA). Instead, a single factor experiment of two alternatives was
executed; the single factor held constant was the file size being
transferred, and the two alternatives compared were the two databuses. An
additional, independent, single-factor experiment was repeated to account for
the other file size, again seeing the databuses as two alternatives. For both
file sizes, each bus alternative was compared using the non-corresponding
measurements approach, finding the difference of the means. The results of
this analysis were confirmed to be statistically significant using a single
factor, two alternative implementation of ANOVA. This entire process was
repeated three more times, with similar outcomes. While this repetition was
not necessary, if nothing else it served as a testament to the
reproducibility of the experimental results. In the process of performing this experiment, it was
discovered that the buffer array size used in the file transfer automation
software also plays a role in the performance of the data transfer, suggesting
possibilities for further experimentation. Additionally, it was discovered
that it is important to know all of the assumptions ahead of time to avoid
intractable surprises due to any oversights. Furthermore, it is good practice
to ensure that the overhead be kept low between time stamps when transferring
the data to reduce sources of error. Finally, it was discovered that
preliminary trials will give insight into the number of trials needed for the
main experiment, and help characterize any “quirkiness” the experimenter
might face during the actual experiment. |