1
application
HP E8491B IEEE-1394-to-VXI
H
HP Application Note
Increasing Test System Throughput
Appendix I. . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Theoretical Performance . . . . . . . . . . . . . . . . . . 13
Performance Factors . . . . . . . . . . . . . . . . . . . . . 13
VXI Contribution. . . . . . . . . . . . . . . . . . . . . . . . . 14
Appendix II . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
HP E8491B Details . . . . . . . . . . . . . . . . . . . . . . . 15
Features of the IEEE 1394 Bus . . . . . . . . . . . . . 16
Other Advantages . . . . . . . . . . . . . . . . . . . . . . . . 16
Table of Contents
2
Part I:
Measuring Real Test I/O Throughput . . . . . . . . 3
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
System Throughput Benchmarks. . . . . . . . . . . . . 3
HP Innovations to FireWire Technology . . . . . . . 3
Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Small Block Measurements. . . . . . . . . . . . . . . . . . 4
High Channel Count . . . . . . . . . . . . . . . . . . . . . . . 5
Functional Test Peek & Poke . . . . . . . . . . . . . . . . 5
Functional Test Scanning . . . . . . . . . . . . . . . . . . . 5
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Part II:
Understanding VXI Slot 0 I/O Benchmarking . 7
Benchmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Configure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Sample or Scan . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Transfer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
System Throughput . . . . . . . . . . . . . . . . . . . . . . . 8
Controller Configurations . . . . . . . . . . . . . . . . . . . 9
GPIB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
FireWire. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
MXI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Embedded VXI Controller . . . . . . . . . . . . . . . . . 12
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
System Throughput Benchmarks
Getting a true objective measurement of
throughput can be difficult. Published
specifications that measure only part of the test
throughput equation don’t tell a complete story.
In throughput testing, the speed of the interface
sending a command to the target device is very
often the only consideration. But by merely
testing the throughput of one element of a real
test, false conclusions can result: other elements
of the test that are critical to throughput are often
ignored, and opportunities to improve throughput
are often missed.
In the real world of test and measurement, the
time to send a command to a working device, take
a reading, and return it for display or output is a
more complete test for system throughput. So
while the test of the technology with infinitely fast
instruments is addressed in this document (see
Appendix I and II), the real measure of I/O
throughput is the system throughput benchmark.
The test cases described in the following pages
illustrate the throughput benchmark for actual
test measurements.
HP Innovations to FireWire Technology
FireWire transmits "packets" of data in big blocks,
much like a local area network (LAN). Because of
the serial nature of data transmission, first-byte
latency—or the time it takes to transfer the first
byte after the command to initiate is given—is an
issue relevant to throughput. Each data packet
carries this overhead of latency, contributing to
the total throughput elapsed time. If each packet
contains one read/write instruction like the typical
GPIB controller, the overhead results in slow
throughput.
The unique big-block aspects of FireWire
technology allow multiple transactions per packet.
This reduces the impact of first-byte latency:
more data can be carried by a given packet,
leading to decreased packet count. HP engineers
developed additional, higher level protocols, and
optimized bit- and byte-ordering, data
transmission, and error detection and correction
of the bit stream. These modifications
contributed to dramatic improvements in real
throughput of IEEE-1394 for test and
measurement applications.
Part I: Measuring Real Test I/O Throughput
3
Introduction
IEEE-1394 ("FireWire") technology is an open,
scalable, flexible, plug-and-play, low-cost serial
interface that is being widely adopted in digital
consumer products and personal computers.
This high-speed serial bus has prompted many
test departments to determine how FireWire can
be used to connect instruments to computers for
test control. Test environments are discovering
that FireWire is ideally suited to test and
measurement applications, allowing significant
reductions in the cost of test.
By moving data faster with large data block
transfers, test time—and consequently the cost
of test—is reduced. Other cost reductions are
achieved in part because serial data transfer
allows the use of a simplified cable design: the
IEEE-1394 high-speed serial bus uses a thin,
flexible, inexpensive cable to provide a fast, easy
connection between computers and external
peripherals. Another cost reduction comes from
simplification of the electronics: the transmitters
and receivers in the standard chip set handle
addressing, initialization, arbitration and protocols,
keeping the cost of IEEE-1394 technology low.
The HP E8491B is Hewlett-Packard’s
implementation of FireWire technology for test
and measurement applications. This IEEE-1394
PC-Link to the VXI interface ships as a C-size,
1-slot, message-based VXI module. The HP
implementation introduces a variety of advantages
as a Slot 0 control solution in the VXI market:
• It is capable of handling the Resource Manager
and Slot 0 responsibilities.
• It can transfer data blocks greater than
64 kilobytes at 14 megabytes per second,
five times faster than its predecessor.
• It eases configuration with the automatic
recognition feature called "hot-plugging,"
which allows the host system to automatically
recognize an IEEE-1394-based device without
powering down the PC.
• It’s scalable, supporting up to 16 mainframes
using daisy chain or star configurations.
• Its data block transfer capability increases
throughput performance well beyond that of
GPIB at a much lower price than MXI, with
results that match embedded solutions.
In terms of architecture, price and speed, FireWire
and embedded solutions are quickly becoming the
dominant choices in Slot 0 control.
4
Test Cases
HP has benchmarked the typical classes of
measurement, from sending the command to
setting up the instrument, to taking the
measurement itself, to the storage or display of
measurements. The benchmarks use typical
instruments to perform the kinds of tests
commonly seen in data acquisition and functional
test applications. The benchmarks involve
configuring real instruments, taking a sample, and
transferring the sample. The Slot 0 controllers are
the HP E8491B FireWire Card, the HP E1406B
GPIB Card, the NI VXI-PCI8015 MXI-2 Card and
the HP E6235A Embedded 200 MHz PC. A 200
MHz HP Kayak XU PC was the external PC. The
programming language used for all the cases was
C++. The operating system used by the
computers was Microsoft
®
Windows NT®.
Small Block Measurements
The first benchmark is for a register-based data
acquisition system sampling various waveforms.
In this test, the test system configures a
HP E1563A digitizer to sample waveforms in
1K, 4K, 8K, 16K and 1M sample sizes. Then the
system transfers the data.
The most interesting of these benchmarks involves
smaller block transfers of 1K, as shown in Chart 1.
IEEE-1394 technology is noted for its throughput
on large blocks; in this benchmark, small block
transfers were also very cost-effective. Actual
measurements for 1K, 8K, 16K and 1M are
available on the HP test and measurement web
site, along with a complete set of benchmark
results. The site is located at
http://www.tmo.hp.com/tmo/techinfo/English/
EFT_ Benchmarks_FireWire.html.
In an application such as antenna test, where
speeds faster than GPIB would improve test time,
a HP E8491B FireWire connection provides a
4X improvement in performance. Yet, this
improvement costs only 10 percent more than
GPIB. Note that MXI-2 provides only a marginal
improvement in throughput over the HP E8491B
at nearly twice the price.
Also note the throughput at a sample size of
8K (Chart 2). The efficiency of the IEEE-1394
technology begins to match that of MXI-2 and
embedded controllers. For a wide variety of
applications, the HP E8491B is the most costeffective solution, even when transferring
relatively small blocks of data.
Chart 1
Chart 2
1K Sample from E1563A
% of Embedded
%
100
80
60
40
20
0
E8491B GPIB
8K Sample from E1563A
% of Embedded
%
100
80
60
40
20
0
E8491B MXI2
GPIB
MXI2
5
High Channel Count
Charts 3 and 4 show results of another solution
where the IEEE-1394 technology provides
increased throughput at a lower cost in a registerbased data acquisition system requiring a large
number of channels and continuous data
acquisition. This solution uses the direct registerbased protocol that communicates at a lower level
than message-based devices.
In this test, a system scans with three E1413C
64-Channel Scanning A/D Converters, a
configuration that provides the volume of channels
required by many of today’s manufacturing lines.
The HP E1406A GPIB module does not transfer
quickly enough to support multiple HP E1413Cs,
whereas IEEE-1394, MXI-2 and embedded PC all
support the configuration. These three options,
while close in throughput, are widely divergent in
price.
The HP E8491B-based solution is the value leader,
providing virtually the same performance as an
embedded PC at a fraction of the cost. The
HP E8491B-based solution is actually faster than
MXI-2 in this application.
Functional Test Peek & Poke
Chart 5 demonstrates the results of "peek and
poke" measurement times in functional test
applications. This test simulates worst-case
results for the IEEE-1394 technology. In this test,
an HP E1330B Digital I/O inputs or outputs a
control signal; a HP E8460A Multiplexer opens
one switch and closes another; a HP E1411B DMM
changes function and range, then takes one
reading. The HP E8491B provides the low cost
solution with a 40 percent improvement over
GPIB. For high speed requirements, the
embedded solution provides an open architecture
with a smaller footprint for manufacturing
applications.
Chart 3
Chart 5
Chart 4
64 Readings @ 100 kHz Rate
% of Embedded
%
120
100
80
60
40
20
0
E8491B GPIB
50K Readings @ 100 kHz Rate
% of Embedded
%
100
80
60
MXI2
40
20
0
E8491B MXI2
GPIB
Register Based Electronic Functional Test
% of Embedded
%
150
100
50
0
E8491B GPIB
MXI2
Loading...