NCS Research Scratchpad
02/20/2004
A few notes from our last meeting:
-
The analytical stability analysis studied the stability of the linearized system, but simulations are for the non-linear system. Will repeat simulation with linearized version.
-
Simulations cut off at 200 ms sampling, but further results seem to exist. Will extend simulations (and analytical analysis) out to approximately 300 ms.
-
Will present a short description of TTP (time-triggered protocol) next time.
-
Will present the sampling period distribution in a form of "survivor plot"
-
Should review Hassibi, et. al. paper, along with papers on jump linear systems.
-
Final thesis should present ideas about effective sampling rate vs. stability.
-
Talked about displaying Fourier transforms for effective sampling rate.
02/10/2004
Papers to review:
-
Neha Gandhi, Joseph L Hellerstein, Sujay S. Parekh, Dawn Tilbury and Yixin Diao. MIMO Control of an Apache Web Server: Modeling and Controller Design, Proceedings of American Control Conference, May 2002 [http://www.research.ibm.com/PM/acc02final2.pdf]
-
Neha Gandhi, Joseph L Hellerstein, Yixin Diao, Sujay S. Parekh and Dawn Tilbury. Using MIMO Feedback Control to Enforce Policies for Interrelated Metrics with Application to the Apache Web Server, Proceedings of NOMS 2002 IEEE/IFIP Network Operatoins and Management Symposium. Piscataway, NJ, IEEE. 2002, p. 219-34., April 2002 [http://www.research.ibm.com/PM/noms2002_MIMO.pdf]
-
Sujay Parekh, Neha Gandhi, Joe Hellerstein, Dawn Tilbury, T. S. Jayram. Using Control Theory to Achieve Service Level Objectives In Performance Management, 2001 [http://www.research.ibm.com/PM/im2001gac.pdf]
-
M. Amirijoo et al., Algorithms for Managing QoS for Real-Time Data Services Using Imprecise Computation, Proc. Of the 9th International Conference on Real-Time Computing Systems and Applications (RTCSA2003), 2003 [http://www.ida.liu.se/~meham/QoSRTDS/publications/files/SNART_extended_abstract.pdf]
-
Timothy X. Brown. Switch Packet Arbitration via Queue-Learning, 2001 [http://www-2.cs.cmu.edu/Groups/NIPS/NIPS2001/papers/psgz/AP11.ps.gz]
-
C.V. Hollot, Vishal Misra, Don Towsley and Wei-Bo Gong. A Control Theoretic Analysis of RED [http://www.ieee-infocom.org/2001/paper/791.pdf]
-
Tao Ye, Shivkumar Kalyanaraman. Adaptive Tuning of RED Using On-line Simulation [http://www.ecse.rpi.edu/Homepages/shivkuma/research/papers/tao-globecom02.pdf]
-
Mikkel Christiansen, Kevin Jeffay, David Ott, F. Donelson Smith. Tuning RED for Web Traffic [http://www.cs.unc.edu/~jeffay/papers/IEEE-ToN-01.ps.gz]
-
Sally Floyd and Van Jacobson. Random Early Detection Gateways for Congestion Avoidance [http://www.icir.org/floyd/papers/early.pdf]
-
Anastasios Stamoulis and Georgios B. Giannakis. Deterministic Time-Varying Packet Fair Queueing for Integrated Services Networks [http://spincom.ece.umn.edu/papers02/jvlsisp02.pdf]
02/06/2004
-
Looked at a website curated by Prof. Liberatore: http://vorlon.cwru.edu/~vxl11/NetBots/. This is sort of an NCS repository, for now. Another NCS repository will be built later.
-
From Simulation 1 of 2/6/2004, we discussed varying a constant delay at the router, rather than varying the sample time. That is, using a discrete controller designed for 50 ms, induce plant sampling every 50 ms, while varying (across simulations) a fixed delay at the controller link from 0 to ?? ms.
-
We discussed the idea of "maximum error first" queueing theory. This is as simple as releasing the next packet from the queue for which the error for the NCS is greatest. I will look for publications from Tilbury, Walsh, and Bushnell regarding this theory.
-
All plots of the delays should use the same time scale, so that they can be compared visually.
-
Plots of delays should omit the initial zero value recorded.
-
We talked about the idea of tracking the "effective sampling rate" for a plant/controller. This can be done with a "sliding window average", with two samples as the base case.
-
I met with Vincenzo Liberatore and his students. Prayas talked about his research, which currently involves RTP (real-time protocol), and extensions to RTP to allow for intelligent retransmission. Prayas has also designed the new NCS repository, which is at http://home.cwru.edu/~pxa32/ncs.
-
Prof. Liberatore brought up the notion of observing the bandwidth utilization at the controller-router link, as a network-centric version of observing the "effective sampling rate" of each plant.
-
Professor Liberatore recommended that delay and drop distributions be plotted using a complimentary community distribution (often called a "survivor plot" in the world of statistics) rather than a histogram, which has the problem of bin size.
-
I need to strictly define what it means to lose control of a plant. When this is strictly defined, (1) the controller can stop sending control signals (thus bounding the error), and instability is easily observed as a binary phenomenon.
01/16/2004
After reading through Wei Zhang's doctoral thesis (http://dora.cwru.edu/msb/pubs/wxzPHD.pdf), I've noticed a couple of open issues that seem to be possible directions for study/research:
-
Decrease of Lyapunov function over multiple sampling periods, rather than just 1
-
Measurements and forward correction for controller-to-actuator delay
-
Revised bounds on transmission error, state error, decrease of Lyapunov function, and transmission period if estimators/correctors are used
-
Online scheduling algorithms
-
Continue the model for delay + dropout
01/06/2004
Some papers/links to read: