poster
PFE - 13
Performance Comparison of EPICS IOC
and MARTe in a Hard Real-Time
Control Application
A.Barbalace1, G. Manduchi1, A. Neto2, G. De Tommasi3, D.F. Valcárcel2, F. Sartori4
1Consorzio
RFX, Associazione EURATOM-ENEA sulla Fusione, Corso Stati Uniti 4, I-35127 Padova, Italy
EURATOM/IST, Instituto de Plasmas e Fusão Nuclear, Av. Rovisco Pais, 1049-001 Lisboa, Portugal
3Associazione EURATOM-ENEA-CREATE, Università di Napoli Federico II, I-80125 Napoli, Italy
4Fusion for Energy, 08019 Barcelona, Spain
2Associação
e-mail: [email protected] , ph. +39 049 8205074, Fax +39 049 8700718
What we have done
EPICS IOC
MARTe
Similarities and Differences
Records
modular components based architectures
GAMs
ReCompile
current configuration is defined in a text file
ReLoading
No Control
computational resource model
Full Control
Case Study
input/output board NI6255 (PXI external rack)
workstation HP Compaq dc 7900 CMT, x86 Intel Core 2 Duo [email protected] GHz, 3 MB L2, 3 GB RAM
Linux version 2.6.29.6 RT patched (rt-24), 1kHz system clock
First Results
Comparing with a Reference Program
Reference
Prog.
min
EPICS
IOC
MARTe
19.0 us
30.0 us
20.9 us
MAX 28.3 us
60.5 us
31.3 us
Avg
39.3 us
25.0 us
22.3 us
min, MAX and average overall IO latency
• the added latency due to MARTe in
respect of the reference program is on
average 2.7 µs.
• the added latency due to EPICS is
17 µs.
1kHz
2kHz
5kHz
10kHz
2%
4%
8%
14%
EPICS IOC 4%
4%
14%
28%
MARTe
7%
14%
29%
Reference
Prog.
4%
CPU load at different clock frequencies
Conclusions
EPICS IOC
MARTe
Adding network queries
Lesson Learned
• MARTe provides a shorter and, above all, more bounded latency;
• EPICS can be used as well for non highly demanding real-time applications.
• Linux with the Real-Time patches appears to be well suited to develop hard Real-Time
applications making it a strong candidate for real-time control in large systems.