Application Note OXAN 1
Customer perception of performance
in IEEE 1394 Applications
Oxford Semiconductor has designed the OXFW911 IEEE
1394 to IDE bridge with improved performance in mind.
Contents
·
·
·
·
·
·
Introduction
Description
They have integrated the 16-bit flash memory onto the
device to minimise the wait states incurred in accessing the
flash. They have also optimised the interaction of the
ARM7-TDMI based hardware and the firmware to reduce
the relative time spent on the SBP-2 protocol conversion.
Optimisation of OXFW911
Graphs of OXFW911 performance
Optimal Performance
Conclusion
Introduction
Graphs of OXFW911 Performance
This application note details the parameters important for
customer perception of performance in IEEE 1394
applications. Much attention is given to the maximum
performance of any given system. However, in the majority
of applications small data block sizes are used and
maximum performance will not be achieved.
Figure 1 details the performance of the OXFW911 with
various block sizes versus the same drive placed on the
internal IDE bus. So far these figures for the OXFW911
have been limited by the speed of the drive used for
testing, and the bus. These figures will be updated once
faster drives are available.
Description
The tests were carried out on a NEC DTLA-307020
20.5GB hard disk using 15-11-2000 firmware. The tests
were carried out on a Windows 2000 system using YAPT.
The Host was a Dell PIII 800 with 128M RAM.
In all data transfer operations, optimal performance is
achieved when large data block sizes are transferred.
Each transaction consists of a protocol stage, a drive spin
stage and a fetch data stage. The larger the amount of
data transferred in each transaction, the smaller
percentage of time is spent on drive spin up and protocol.
Optimal Performance
While testing the OXFW911, it has been noted that several
factors influence the maximum speed obtained. The main
factors are the operating system used, the drive speed and
the OHCI card.
In normal operation, the most frequently used applications
do not transfer large amounts of data. An example of this
is browsing a directory structure. This can be done
frequently and transfers, at most, 4k of data per
transaction.
Conclusion
Customer perception of drive performance is determined
during normal use. In most cases the data block sizes
transferred are small. Therefore, increased performance at
small block sizes will greatly increase the customers’
perception of drive speed.
Although all companies quote the highest data rate
achievable, the customer will rarely see this performance.
Their judgement of speed will be based on the speed
during normal, small block size performance.
This area has become of concern within the IEEE 1394
community. Protocol changes are being considered to
reduce the time spent on the protocol phase of the
transaction and so greatly improve small block size
performance.
The OXFW911 has been optimised in both hardware and
firmware to take full advantage of the ARM7TDMI
processor. This optimisation has resulted in excellent
performance at small block sizes, this can be better then
when the drive is connected internally.
.
Optimisation of OXFW911
Oxford Semiconductor Ltd.
25 Milton Park, Abingdon, Oxon, OX14 4SH, UK
Tel: +44 (0)1235 824900
Ó Oxford Semiconductor 2000.
OXAN 1 Application Note Revision 1 – December 2000
Part Numbers – OXFW911
Fax: +44(0)1235 821141
