![](bg9.png)
Figure 7: Experimental 99
th
Percentile Response Times
(Markers) for Different Loads (LD) and Load Ratios (LDR)
at Two Different Network Bandwidth Ratios (BWR), Com-
pared with the Theoretical Expectations of Eq.(4) (Dashed
Lines).
lighted some limitations of the model arising when
getting closer to the instability region (saturation of
the reserved/available computational or networking
bandwidth), in addition to the well-known limitation
due to non-negligible scheduling overheads as hap-
pening with too small CPU reservation periods.
This shows that real-time containers really enable
a predictable performance for the hosted software
components, so that we can build abstract, high-level
performance models useful for designing applications
with strong end-to-end QoS guarantees.
As a future work, the model validation presented
in this paper can be extended in various ways. First,
scenarios with more concurrent clients could be taken
into consideration. This would also require some mi-
nor modifications to the performance model.
Second, the isolation capabilities of our proposed
architecture has to be validated under more complex
interference scenarios with a multitude of workload
types. For example, mechanisms to control storage
access and its associated model could be added – at
least when using SSD drives.
Third, the studied model considers containers us-
ing only a single CPU core. However, in many NFV
scenarios the containers run concurrent servers us-
ing multiple threads for handling the clients requests.
Hence, leveraging multiple CPU cores per container
would be more realistic. This is among the planned
extensions of this work in our future investigations.
Finally, the performance of some real virtualized
network function could be analyzed – e.g., deploy-
ing Open Air Interface
10
, Kamailio
11
or similar open-
source software.
REFERENCES
Abeni, L., Balsini, A., and Cucinotta, T. (2018). Container-
based real-time scheduling in the linux kernel. In Pro-
ceedings of the Embedded Operating System Work-
shop 2018 (EWiLi’18), Torino, Italy.
Aditya, P., Akkus, I. E., Beck, A., Chen, R., Hilt, V., Ri-
mac, I., Satzke, K., and Stein, M. (2019). Will server-
less computing revolutionize nfv? Proceedings of the
IEEE, 107(4):667–678.
Akkus, I. E., Chen, R., Rimac, I., Stein, M., Satzke, K.,
Beck, A., Aditya, P., and Hilt, V. (2018). SAND:
Towards high-performance serverless computing. In
2018 USENIX Annual Technical Conference (USENIX
ATC 18), pages 923–935, Boston, MA. USENIX As-
sociation.
Ali-Eldin, A., Tordsson, J., and Elmroth, E. (2012). An
adaptive hybrid elasticity controller for cloud infras-
tructures. In 2012 IEEE Network Operations and
Management Symposium, pages 204–212.
Allen, A. O. (1978). Probability, Statistics, and Queue-
ing Theory with Computer Science Applications. Aca-
demic Press, Inc., Orlando, FL, USA.
Ardagna, C. A., Asal, R., Damiani, E., and Vu, Q. H.
(2015). From security to assurance in the cloud: A
survey. ACM Comput. Surv., 48(1):2:1–2:50.
Armbrust, M., Fox, A., Griffith, R., Joseph, A. D., Katz,
R., Konwinski, A., Lee, G., Patterson, D., Rabkin, A.,
Stoica, I., and Zaharia, M. (2010). A view of cloud
computing. Commun. ACM, 53(4):50–58.
Cucinotta, T., Abeni, L., Marinoni, M., Balsini, A., and
Vitucci, C. (2019). Reducing temporal interference
in private clouds through real-time containers. In
Proceedings of the IEEE International Conference on
Edge Computing, Milan, Italy. IEEE.
Cucinotta, T., Anastasi, G., and Abeni, L. (2009). Respect-
ing temporal constraints in virtualised services. In
2009 33rd Annual IEEE International Computer Soft-
ware and Applications Conference, volume 2.
Cucinotta, T., Checconi, F., Zlatev, Z., Papay, J., Boni-
face, M., Kousiouris, G., Kyriazis, D., Varvarigou,
10
More information can be found at: https://www.
openairinterface.org/.
11
More information can be found at: https://www.
kamailio.org/w/.
Performance Modeling in Predictable Cloud Computing
77