Hi,
We were performing some experiments in NS-2 using the IEEE 802.16 / WiMAX
mesh patch (http://cng1.iet.unipi.it/wiki/index.php/Ns2mesh80216) on tuning
the holdoff exponents for distributed coordinated scheduler in IEEE 802.16
mesh.
*Setup: *
We considered a chain of odd number of nodes.
For example, a chain of 7 nodes looks like:
n1 - n2 - n3 - n4 - n5 - n6 - n7
*Holdoff Time: *
The holdoff time (in slots) of a node is the number of slots it has to wait
between two successive control channel access.
The expected holdoff time E[t_k] of node k is given by:
E[t_k] = 2^(base_k + x_k) + E[s_k]
where base_k is the holdoff time base, x_k is the holdoff exponent and
E[s_k] is the expected number of slots in which node k fails before it wins
the competition for control channel access.
The above relation follows from the following paper:
*M. Cao, W. Ma, Q Zhang, X. Wang and W. Zhu, “Modeling and Performance
Analysis of the Distributed Scheduler in IEEE 802.16 Mesh Mode,” Proc. 6th
ACM MobiHoc, Urbana-Champaign, May 2005
*
*Method to reduce holdoff time:*
Now for a linear chain, it is useful to set
base_k = floor(ln (N_k))
where N_K is the number of two-hop neighbors of node k.
This idea is suggested in
S.Y. Wang, C.C. Lin, H. Chu and T.W. Hsu, “Improving the Performances of
Distributed Coordinated Scheduling in IEEE 802.16 Mesh Networks,” IEEE
Transactions on Vehicular Technology, 57(4): 2531-2547, July 2008
In the above paper, Wang et al. also suggested that hold off exponents
should be set to zero.
Thus for the chain of 7 nodes, holdoff bases are {1, 1, 2, 2, 2, 1, 1} and
the holdoff exponents are {0, 0, 0, 0, 0, 0, 0}.
*Our observations:*
For the chain of 7 nodes, we used the same values for holdoff bases, that
is, {1, 1, 2, 2, 2, 1, 1} but used the set {0, 0, 1, 0, 1, 0, 0} for the
exponents.
Note: There is a provision to set the holdoff exponents manually via a tcl
script in the patch mentioned above.
Then we carried out all the experiments reported in the above paper by Wang
et al., that is, we measured TCP throughput, UDP throughput and round trip
time (RTT).
We found that our setting produced slightly better results.
Similar observations were made for longer chains of odd lengths when the
exponents of the third node from the left and the third node from the right
are set to one and all other exponents are set to zero.
*This observation appears very surprising since the delay is supposed to
increase for larger exponents.*
Can anyone please throw some light on this?
Your feedback is appreciated.
Regards,
Debarshi
Regards,
Debarshi
