STAR-GOAT

Strategic Tester for Aerospace Research · Ground Operation Acceptance Testing

Performance datasheet

Measured throughput, timing accuracy and reaction time

Document: Performance datasheet
Revision: 1.0
Measured: December 12, 2025
Published: July 29, 2026
Contact: star-goat@jki.net

Executive summary

Headline results

STAR-GOAT is a complete test stand platform: high-speed data acquisition, deterministic real-time control, and operator visualization in one configurable system.

This performance datasheet provides validated benchmarking data for STAR-GOAT, JKI's complete data acquisition and control software for mission-critical test stands. The measurements cover high-speed I/O monitoring, real-time control, and multi-client remote visualization across three hardware platforms and a range of system configurations.

100 µs Worst sequence timing error at 82% CPU load on RT PXIe-8881

Flat 8% CPU load from 16 to 480 kHz per channel on DAQmx acquisition

0.25% Average CPU overhead per additional remote client

130x Better sequence timing accuracy on RT than Windows at baseline, 61 µs against 8 ms

61 µs max timing error, RT baseline 1.04 ms sequence reaction time, RT baseline 836 DAQmx channels at 16000 Hz for 8.0% CPU 10 concurrent operator stations 3.86 MS/s aggregate, PXI-6375 card limit 8 ms Windows timing accuracy, Xeon Gold workstation

Test configuration

What was tested, and on what

Performance testing used a baseline configuration approach to enable meaningful platform comparisons and scalability analysis. A representative test stand configuration was established for each platform, incorporating realistic channel counts, data rates, and control requirements.

Dimensions characterized

From the baseline profile, systematic testing characterized performance across four dimensions.

Channel scalability. DAQmx, OPC and calculation channel counts varied independently.
Data rate. Acquisition rates increased from 16 kHz to 3.84 MHz.
Multi-client loading. 1–10 concurrent remote HMI connections.
Real-time determinism. Sequence timing accuracy and reaction time under CPU load.

All measurements captured CPU utilization, timing accuracy, and system responsiveness.

Hardware under test

Controllers and acquisition hardware under test
RT platform	PXIe-8881 real-time controller
Windows platform	Intel Xeon Gold 6226R at 2.90 GHz, Windows 11 Pro for Workstations
cRIO platform	cRIO-9048 real-time controller
Acquisition cards	Multiple PXI-6375 cards. Baseline channel counts exceed a single card's aggregate budget and are distributed across cards.

Scope of the channel load sweeps

Channel load performance tests were executed on the RT PXIe-8881. Similar trends may be inferred for the cRIO and Windows PC platforms.

Results by platform

Baseline profiles, three platforms

Baseline profiles were generated for three different platforms: RT PXIe-8881, RT cRIO-9048, and a high-performance Windows PC. Later sections push individual performance vectors, such as channel count and data rate, to characterize the impact on STAR-GOAT performance.

Baseline configuration and measured result, per platform
Platform	Baseline test configuration					Measured result
Platform	DAQmx channels	DAQmx data rate	OPC channels	Calc channels	Control loops	CPU	Sequence timing accuracy	Sequence reaction time
RT PXIe-8881	836	16000 Hz	40	120	5	8.0%	61 µs	1.04 ms
Windows PC	128	2000 Hz	100	200	5	15%	8 ms	10 ms
cRIO-9048	168	2000 Hz	100	10	1	45%	12 ms	18.1 ms

Read the Windows numbers with this

Windows PC benchmarks were performed on a high-performance server: Intel Xeon Gold 6226R at 2.90 GHz running Windows 11 Pro for Workstations. Metrics such as sequence reaction time may differ on lower-performance machines.

Typical Windows PCs may have reaction times closer to 100–200 ms.

Channel load

Channel scalability and CPU cost

Scalability and CPU utilization by channel type. These benchmarks were performed on the RT PXIe-8881, manipulating the baseline configuration to add channels and increase data rates to characterize system performance at varying channel loads.

DAQmx data rate against CPU load

Figure 1CPU load against DAQmx data rate per channel. RT PXIe-8881, log scale. Straight segments join measured points, no interpolation.

Figure 1 measured points
Data rate	CPU load
16 kHz	8.0%
32 kHz	8.0%
120 kHz	8.0%
480 kHz	8.0%
1920 kHz	10.0%
3840 kHz	12.0%

The 3.86 MS/s ceiling belongs to the acquisition card

In order to increase data rate beyond 16 kHz with the PXI-6375, total channel count was reduced to maintain a maximum of 3.86 MS/s aggregate sample rate. That 3.86 MS/s figure is the aggregate sample rate limit of the PXI-6375 card used in this test. It is not a STAR-GOAT limit, and it changes with the acquisition hardware.

Channel count against CPU load

Figure 2CPU load against OPC channel count at 50 Hz. RT PXIe-8881.

Figure 2 measured points
OPC channels	CPU load
40	8.0%
80	8.0%
160	8.0%
320	11.0%
640	13.0%

Figure 3CPU load against calculation channel count at 500–1000 Hz. RT PXIe-8881.

Figure 3 measured points
Calc channels	CPU load
40	8.0%
100	12.0%
200	22.0%
400	38.0%
800	75.0%

Calculation loop rate

Calculations were configured to run at 1 kHz. Actual loop rate decreased to 500 Hz starting at 200 calculation channels.

Bench against field

Bench-characterized to 1000 channels. Deployed in the field beyond 2500 channels, predominantly OPC UA.

Operator stations

Cost of concurrent clients

System performance with multiple remote clients connected. A test stand typically runs several operator stations at once because different engineers own different subsystems.

Concurrent operator stations against CPU load

Figure 4CPU load against concurrent operator stations, 1 to 10 clients. RT PXIe-8881.

Summary

Average 0.25% CPU overhead per additional client.

Figure 4 measured points

Clients 1 to 5
Clients	CPU load
1	6.8%
2	7.0%
3	7.25%
4	7.5%
5	7.75%

Clients 6 to 10
Clients	CPU load
6	8.0%
7	8.25%
8	8.5%
9	8.75%
10	9.25%

Sequencer

Sequence timing accuracy and reaction time

Real-time sequence control accuracy and response times across platforms and system loads.

Platform performance overview
Platform	Sequence accuracy range	Sequence reaction time range	HWTSP available	Performance rating
PXIe-8881 (RT)	61–100 µs max error	1.04–2.15 ms average	Yes	Excellent
Windows PC	2.43–122 ms max error	2.21–18.2 ms average	Yes	Good
cRIO-9048	9.5–12 ms max error	18.1–36.7 ms average	No	Good

Sequence accuracy: Maximum deviation between actual and configured sequence timing.
Sequence reaction time: Response time to external events during sequence execution.
HWTSP, hardware timed single point: Low-latency response mode for critical timing applications.

RT PXIe-8881 against CPU load

Figure 5Maximum timing error and reaction time against CPU load. RT PXIe-8881, left axis microseconds, right axis milliseconds.

Figure 5 measured points
CPU load	Max timing error	Reaction time
7.56%	61 µs	1.04 ms
13.43%	73 µs	1.25 ms
21.25%	82 µs	1.1 ms
39%	71 µs	1.2 ms
58.87%	69 µs	1.95 ms
82.17%	100 µs	2.15 ms

Windows PC and cRIO-9048 against CPU load

Figure 6Maximum timing error and reaction time against CPU load. Windows PC, Xeon Gold 6226R.

Figure 6 measured points, maximum timing error
CPU load	Max timing error
9.57%	2.43 ms
11.51%	2 ms
15.86%	9.07 ms
18.3%	24.5 ms
19.95%	48.5 ms
32.83%	122 ms

Figure 6 measured points, reaction time
CPU load	Reaction time
10%	7.1 ms
11.51%	2.21 ms
15%	9.95 ms
18.3%	14.4 ms
19.95%	17 ms
32.83%	18.2 ms

Figure 7Maximum timing error and reaction time against CPU load. cRIO-9048, two measured load points.

Figure 7 measured points
CPU load	Max timing error	Reaction time
44.89%	12 ms	18.1 ms
62.56%	9.5 ms	36.7 ms

Axis ranges

CPU load ranges differ per platform because each platform was swept over the load range its baseline reaches. Compare the values, not the plot shapes.

Best measured platform: PXIe-8881 holds maximum timing error at or under 100 µs even at 82% CPU load.
Real-time advantage: RT accuracy stays inside 100 µs across the full load sweep. Windows ranges from 2 ms to 122 ms over a 23-point CPU swing.
HWTSP performance: 1.04 ms average reaction time on PXIe-8881.
Load tolerance: PXIe-8881 was swept to 82% CPU load with no loss of timing accuracy.

Configuration guidelines

Reading these numbers into a system design

Each line below restates a measurement from the preceding sections, in the form most useful when sizing a controller.

Design questions against measured results
Design question	What the measurements show
Which channel type costs most CPU	Calculation channels. On the RT PXIe-8881, 800 calculation channels at 500–1000 Hz reached 75% CPU, while 640 OPC channels at 50 Hz reached 13% and DAQmx acquisition stayed flat at 8% from 16 to 480 kHz per channel.
Calculation loop rate under load	Configured at 1 kHz. Actual loop rate decreased to 500 Hz starting at 200 calculation channels.
Aggregate sample rate ceiling	Set by the acquisition card, 3.86 MS/s on the PXI-6375 used here. Channel count and per-channel rate trade against each other within that budget.
Where determinism figures apply	The real-time controllers. On the RT PXIe-8881, maximum sequence timing error measured between 61 µs and 100 µs across 7.56% to 82.17% CPU load. Windows measured 8 ms accuracy and 10 ms reaction time on a Xeon Gold workstation, with typical machines closer to 100–200 ms reaction time.
HWTSP availability	Available on the PXIe-8881 and Windows platforms in these tests. Not available on the cRIO-9048.
Operator station budget	0.25% average CPU per additional remote client, measured to 10 concurrent clients.
Applicability of channel load trends	Characterized on the RT PXIe-8881. Similar trends may be inferred for the cRIO and Windows PC platforms.

Bench-characterized to 1000 channels Field deployments beyond 2500 channels 10 concurrent operator stations measured 3 controller platforms