In this chapter we consider multirate loss models of random arriving calls not only with elastic bandwidth requirements upon arrival but also elastic bandwidth allocation during service. Calls may retry several times upon arrival, requiring less bandwidth each time, in order to be accepted for service. If call admission is not possible with the last (least) bandwidth requirement then bandwidth compression is attempted.
In the elastic single‐retry model (E‐SRM), we consider a link of capacity b.u. that accommodates elastic calls of service‐classes. Calls of service‐class follow a Poisson process with mean arrival rate and have a peak‐bandwidth requirement of b.u. and an exponentially distributed service time with mean . To introduce bandwidth compression, we permit the occupied link bandwidth to virtually exceed up to a limit of b.u. Let be the occupied link bandwidth, , when a new service‐class call arrives in the link. Then, for call admission we consider the following cases:
Similar to the E‐EMLM, when a service‐class call, with bandwidth (or ), departs from the system, the remaining in‐service calls of each service‐class expand their bandwidth in proportion to their initially assigned bandwidth (or ). After bandwidth compression/expansion, elastic service‐class calls increase/decrease their service time so that the product service time by bandwidth remains constant.
Similar to the SRM, the steady state probabilities in the E‐SRM do not have a PFS, since LB is destroyed between adjacent states (see Figure 4.1). Thus, the unnormalized values of can be determined by an approximate but recursive formula, as presented in Section 4.1.2.
Table 4.1 The state space and the occupied link bandwidth (Example 4.1).
(before compr.) | (after compr.) | (before compr.) | (after compr.) | ||||||||
0 | 0 | 0 | 1.00 | 0 | 0 | 2 | 0 | 0 | 1.00 | 2 | 2 |
0 | 0 | 1 | 1.00 | 2 | 2 | 2 | 0 | 1 | 1.00 | 4 | 4 |
0 | 0 | 2 | 1.00 | 4 | 4 | 2 | 0 | 2 | 0.67 | 6 | 4 |
0 | 0 | 3 | 0.67 | 6 | 4 | 2 | 1 | 0 | 0.80 | 5 | 4 |
0 | 1 | 0 | 1.00 | 3 | 3 | 3 | 0 | 0 | 1.00 | 3 | 3 |
0 | 1 | 1 | 0.80 | 5 | 4 | 3 | 0 | 1 | 0.80 | 5 | 4 |
1 | 0 | 0 | 1.00 | 1 | 1 | 3 | 1 | 0 | 0.67 | 6 | 4 |
1 | 0 | 1 | 1.00 | 3 | 3 | 4 | 0 | 0 | 1.00 | 4 | 4 |
1 | 0 | 2 | 0.80 | 5 | 4 | 4 | 0 | 1 | 0.67 | 6 | 4 |
1 | 1 | 0 | 1.00 | 4 | 4 | 5 | 0 | 0 | 0.80 | 5 | 4 |
1 | 1 | 1 | 0.67 | 6 | 4 | 6 | 0 | 0 | 0.67 | 6 | 4 |
To facilitate the recursive calculation of , we replace by the state‐dependent compression factors per service‐class , which have a similar role to and have already been described in the E‐EMLM. The only difference is that apart from , which are given by (3.8), we should also define to account for retry calls of service‐class in state . The form of is the following [1]:
where , , and
Consider again Example 4.1 (, , ):
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with the exact 0.1706) |
The of service‐class 2 calls, , when they require b.u. upon arrival is given by:
(compare with the exact 0.7821) |
The CBP of service‐class 2, , is given by:
(compare with the exact 0.3325) |
The link utilization is determined by:
b.u. (the same as the exact value). |
Table 4.2 The values of the state‐dependent compression factors and (Example 4.2).
0 | 0 | 0 | 1.0 | 1.00 | 1.00 | 1.00 | 2 | 0 | 0 | 1.0 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 1.0 | 1.00 | 1.00 | 1.00 | 2 | 0 | 1 | 1.0 | 1.00 | 1.00 | 1.00 |
0 | 0 | 2 | 1.0 | 1.00 | 1.00 | 1.00 | 2 | 0 | 2 | 1.625 | 0.7692 | 0.00 | 0.6154 |
0 | 0 | 3 | 1.5 | 0.00 | 0.00 | 0.6667 | 2 | 1 | 0 | 1.25 | 0.80 | 0.80 | 0.00 |
0 | 1 | 0 | 1.0 | 1.00 | 1.00 | 1.00 | 3 | 0 | 0 | 1.0 | 1.00 | 1.00 | 1.00 |
0 | 1 | 1 | 1.25 | 0.00 | 0.80 | 0.80 | 3 | 0 | 1 | 1.25 | 0.80 | 0.00 | 0.80 |
1 | 0 | 0 | 1.0 | 1.00 | 1.00 | 1.00 | 3 | 1 | 0 | 1.6875 | 0.7407 | 0.5926 | 0.00 |
1 | 0 | 1 | 1.0 | 1.00 | 1.00 | 1.00 | 4 | 0 | 0 | 1.0 | 1.00 | 1.00 | 1.00 |
1 | 0 | 2 | 1.25 | 0.80 | 0.00 | 0.80 | 4 | 0 | 1 | 1.75 | 0.7143 | 0.00 | 0.5714 |
1 | 1 | 0 | 1.0 | 1.00 | 1.00 | 1.00 | 5 | 0 | 0 | 1.25 | 0.80 | 0.00 | 0.00 |
1 | 1 | 1 | 1.5625 | 0.80 | 0.64 | 0.64 | 6 | 0 | 0 | 1.875 | 0.6667 | 0.00 | 0.00 |
To describe the analytical model in the steady state, we consider a link of capacity b.u. that accommodates calls of two service‐classes with traffic parameters: () for service‐class 1 and () for service‐class 2. Calls of service‐class 2 have retry parameters with and . Let be the virtual capacity so that the maximum permitted bandwidth compression is for calls of both service‐classes.
Although the E‐SRM is a non‐PFS model, we will use the LB of (3.11), initially for calls of service‐class 1:
where with , and
Based on (4.4) and multiplying both sides of (4.3) with , we have:
where and the values of are given by (4.2).
Based on the CAC of the E‐SRM, we consider the following LB equations for calls of service‐class 2:
where with and
Based on (4.7) and multiplying both sides of (4.6) with , we have:
where and the values of are given by ( 4.2).
where and
Based on (4.10) and multiplying both sides of (4.9) with , we have:
where and the values of are given by ( 4.2).
Equations (4.5), (4.8), and (4.11) lead to the following system of equations:
Equations (4.12)–(4.14) can be combined into one equation by assuming that calls with are negligible when and calls with are negligible when :
where for , otherwise and for , otherwise .
Note that the approximations introduced in (4.15) are similar to those introduced in the STM of [2].
Since , when , it is proved in [ 2] (see also (2.9)) that:
for and for , otherwise .
Reminder: To prove (4.16), the migration approximation is needed, which assumes that the population of retry calls of service‐class 2 is negligible in states .
When and based on ( 4.2), ( 4.15) can be written as:
To introduce the link occupancy distribution q(j) in (4.17), we sum both sides of ( 4.17) over the set of states :
Since by definition , (4.18) is written as:
where for .
The combination of ( 4.16) and (4.19) gives the following approximate recursive formula for the calculation of in the case of two service‐classes, when only calls of service‐class 2 have retry parameters (for ):
where for , otherwise , and for , otherwise .
In the case of service‐classes and assuming that all service‐classes may have retry parameters (4.20) takes the general form:
where , .
Having determined the unnormalized values of , we can calculate [1]:
Note that if , then refers to the and the summation in (4.23) should start from .
Consider again Example 4.1 (, , ).
The normalization constant is: .
The state probabilities are:
(compare with the exact 0.1706) |
(compare with the exact 0.7821) |
(compare with the exact 0.3325) |
(compare with the exact 0.4251) |
It is apparent that even in small E‐SRM examples the error introduced by the assumption of LB, the introduction of and , and the migration approximation is not significant.
We now consider the E‐SRM under the BR policy (E‐SRM/BR) with BR parameter for service‐class calls (). For CAC in the E‐SRM/BR, we consider the following cases:
As far as the values of , , and are concerned they are determined by (3.8), (4.1), and ( 4.2), respectively.
Consider again Example 4.1 (, , ) and apply the BR parameters b.u. and b.u. to calls of service‐class 1 and 2, respectively.
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with 0.1706 in the E‐SRM) |
The of service‐class 2 calls, , is given by:
(compare with 0.7821 in the E‐SRM) |
Due to the selection of the BR parameters, the CBP of service‐class 2, equals :
(compare with 0.3325 in the E‐SRM) |
The link utilization is determined by:
b.u. (compare with 2.832 in the E‐SRM) |
The solution of this linear system is:
Based on the values of , we determine the values of :
Based on the values of , we obtain the following CBP:
(compare with the exact 0.30737) |
The of service‐class 2 calls, , is given by:
(compare with the exact 0.77264) |
The CBP of service‐class 2, , equals :
The link utilization is determined by:
b.u. (compare with the exact 2.771) |
In the E‐SRM/BR, the recursive calculation of is based on the Roberts method (see Section 1.3.2.2), which leads to the formula [4]:
where and .
Based on (4.29), the following performance measures can be calculated:
Note that if , then refers to the and the summation in (4.31) should start from .
Consider again Example 4.4 (, , ).
The normalization constant is:
. |
The state probabilities are:
(compare with the exact 0.30737) |
(compare with the exact 0.77264) |
(compare with the exact 0.30737) |
(compare with the exact 0.39782) |
It is apparent that even in small E‐SRM/BR examples the error introduced by the assumption of LB, the introduction of and , the migration approximation, and the application of the BR policy remains acceptable.
Similar to the MRM, in the elastic multi‐retry model (E‐MRM) a blocked call of service‐class can have more than one retry parameter for , where and .
To simply describe the CAC, we assume that a service‐class call has a peak‐bandwidth requirement of b.u. and may retry twice to be connected in the system, the first time with and the second time (if blocked with ) with . Then, for call admission, we consider the following cases:
Similar to the E‐SRM, when a service‐class call, with bandwidth (or or ), departs from the system, the remaining in‐service calls of each service‐class expand their bandwidth in proportion to their initially assigned bandwidth (or or ). After bandwidth compression/expansion, elastic service‐class calls increase/decrease their service time so that the product service time by bandwidth remains constant.
Similar to the E‐SRM, the steady state probabilities in the E‐MRM do not have a PFS. Thus, the unnormalized values of can be determined by an approximate but recursive formula, as presented in Section 4.3.2.
Consider again Example 2.6 (, , ) and let b.u.
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with 0.545687 in the MRM) |
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
(compare with 0.876443 in the MRM) |
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
(compare with 0.725882 in the MRM) |
The CBP of service‐class 2, , refers to service‐class 2 retry calls which require b.u. and is given by:
(compare with 0.545687 in the MRM) |
The link utilization is determined by:
b.u. (compare with 2.148 in the MRM) |
Table 4.3 The state space and the occupied link bandwidth (Example 4.6).
(before compr.) | (after compr.) | (before compr.) | (after compr.) | ||||||||||
0 | 0 | 0 | 0 | 1.00 | 0 | 0 | 1 | 0 | 0 | 2 | 1.00 | 3 | 3 |
0 | 0 | 0 | 1 | 1.00 | 1 | 1 | 1 | 0 | 0 | 3 | 0.75 | 4 | 3 |
0 | 0 | 0 | 2 | 1.00 | 2 | 2 | 1 | 0 | 1 | 0 | 1.00 | 3 | 3 |
0 | 0 | 0 | 3 | 1.00 | 3 | 3 | 1 | 0 | 1 | 1 | 0.75 | 4 | 3 |
0 | 0 | 0 | 4 | 0.75 | 4 | 3 | 1 | 1 | 0 | 0 | 0.75 | 4 | 3 |
0 | 0 | 1 | 0 | 1.00 | 2 | 2 | 2 | 0 | 0 | 0 | 1.00 | 2 | 2 |
0 | 0 | 1 | 1 | 1.00 | 3 | 3 | 2 | 0 | 0 | 1 | 1.00 | 3 | 3 |
0 | 0 | 1 | 2 | 0.75 | 4 | 3 | 2 | 0 | 0 | 2 | 0.75 | 4 | 3 |
0 | 1 | 0 | 0 | 1.00 | 3 | 3 | 2 | 0 | 1 | 0 | 0.75 | 4 | 3 |
0 | 1 | 0 | 1 | 0.75 | 4 | 3 | 3 | 0 | 0 | 0 | 1.00 | 3 | 3 |
1 | 0 | 0 | 0 | 1.00 | 1 | 1 | 3 | 0 | 0 | 1 | 0.75 | 4 | 3 |
1 | 0 | 0 | 1 | 1.00 | 2 | 2 | 4 | 0 | 0 | 0 | 0.75 | 4 | 3 |
To facilitate the recursive calculation of , we replace by the state‐dependent compression factors per service‐class , , and , . The values of are given by (3.8), while those of are determined by:
where , , , and
Consider again Example 4.6 (, , ).
Table 4.4 The values of the state‐dependent compression factors and (Example 4.7).
0 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 3 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 4 | 1.3333 | 0.00 | 0.00 | 0.00 | 0.75 |
0 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 2 | 1.3333 | 0.00 | 0.00 | 0.75 | 0.75 |
0 | 1 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 1 | 0 | 1 | 1.3333 | 0.00 | 0.75 | 0.00 | 0.75 |
1 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 3 | 1.3333 | 0.75 | 0.00 | 0.00 | 0.75 |
1 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 1 | 1 | 1.3333 | 0.75 | 0.00 | 0.75 | 0.75 |
1 | 1 | 0 | 0 | 1.3333 | 0.75 | 0.75 | 0.00 | 0.00 |
2 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 2 | 1.3333 | 0.75 | 0.00 | 0.00 | 0.75 |
2 | 0 | 1 | 0 | 1.3333 | 0.75 | 0.00 | 0.75 | 0.00 |
3 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
3 | 0 | 0 | 1 | 1.3333 | 0.75 | 0.00 | 0.00 | 0.75 |
4 | 0 | 0 | 0 | 1.3333 | 0.75 | 0.00 | 0.00 | 0.00 |
Following the analysis of Section 4.1.2.1, the calculation of the unnormalized values of is based on an approximate but recursive formula whose proof is similar to that of ( 4.21) [ 1]:
where , and .
Having determined the unnormalized values of via (4.35) we can calculate [ 1]:
Consider again Example 4.6 (, , ):
The normalization constant is:
. |
The state probabilities are:
(compare with the exact 0.470996) |
(compare with the exact 0.971428) |
(compare with the exact 0.893645) |
(compare with the exact 0.470996) |
(compare with the exact 0.48485) |
It is apparent that even in small E‐MRM examples the error introduced by the assumption of LB, the introduction of , , and the migration approximation is not significant.
Compared to the E‐SRM/BR, in the elastic multi‐retry model under the BR policy (E‐MRM/BR) with BR parameter for service‐class calls (), blocked calls of service‐class can retry more than once to be connected in the system.
Consider again Example 4.6 (, , ) and let the BR parameters and , so that .
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with 0.819672 in the MRM/BR) |
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
. |
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
(compare with 0.606557 in the MRM/BR) |
The CBP of service‐class 2, , refers to service‐class 2 retry calls which require b.u. and is given by:
(compare with 0.514754 in the MRM/BR) |
The link utilization is determined by:
b.u. (compare with 1.941 in the MRM/BR) |
Table 4.5 The state space and the occupied link bandwidth (Example 4.9).
(before compr.) | (after compr.) | (before compr.) | (after compr.) | ||||||||||
0 | 0 | 0 | 0 | 1.00 | 0 | 0 | 1 | 0 | 0 | 0 | 1.00 | 1 | 1 |
0 | 0 | 0 | 1 | 1.00 | 1 | 1 | 1 | 0 | 0 | 1 | 1.00 | 2 | 2 |
0 | 0 | 0 | 2 | 1.00 | 2 | 2 | 1 | 0 | 0 | 2 | 1.00 | 3 | 3 |
0 | 0 | 0 | 3 | 1.00 | 3 | 3 | 1 | 0 | 0 | 3 | 0.75 | 4 | 3 |
0 | 0 | 0 | 4 | 0.75 | 4 | 3 | 1 | 0 | 1 | 0 | 1.00 | 3 | 3 |
0 | 0 | 1 | 0 | 1.00 | 2 | 2 | 1 | 0 | 1 | 1 | 0.75 | 4 | 3 |
0 | 0 | 1 | 1 | 1.00 | 3 | 3 | 2 | 0 | 0 | 0 | 1.00 | 2 | 2 |
0 | 0 | 1 | 2 | 0.75 | 4 | 3 | 2 | 0 | 0 | 1 | 1.00 | 3 | 3 |
0 | 1 | 0 | 0 | 1.00 | 3 | 3 | 2 | 0 | 0 | 2 | 0.75 | 4 | 3 |
0 | 1 | 0 | 1 | 0.75 | 4 | 3 |
To facilitate the recursive calculation of in the E‐MRM/BR, we replace by the state‐dependent compression factors per service‐class , and , . The values of and are given by (3.8) and (4.33), respectively.
Consider again Example 4.9 (, , ):
Table 4.6 The values of the state‐dependent compression factors and (Example 4.10).
0 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 3 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 4 | 1.3333 | 0.00 | 0.00 | 0.00 | 0.75 |
0 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 2 | 1.3333 | 0.00 | 0.00 | 0.75 | 0.75 |
0 | 1 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 1 | 0 | 1 | 1.3333 | 0.00 | 0.75 | 0.00 | 0.75 |
1 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 3 | 1.3333 | 0.75 | 0.00 | 0.00 | 0.75 |
1 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 1 | 1 | 1.3333 | 0.75 | 0.00 | 0.75 | 0.75 |
2 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 2 | 1.3333 | 0.75 | 0.00 | 0.00 | 0.75 |
Following the analysis of Section 4.2.2.1, the calculation of the unnormalized values of is based on an approximate but recursive formula whose proof is similar to that of ( 4.29) [ 4]:
where , , ,
Having determined the unnormalized values of via (4.40) we can calculate:
Consider again Example 4.9 (, , ):
The normalization constant is:
. |
The state probabilities are:
(compare with the exact 0.882627) |
(compare with the exact 0.969125) |
(compare with the exact 0.882627) |
(compare with the exact 0.385833) |
(compare with the exact 0.39813) |
It is apparent that in small E‐MRM/BR examples the error introduced by the assumption of LB, the introduction of and , the migration approximation, and the application of the BR policy can be acceptable.
Consider a single link of capacity b.u. that accommodates three service‐classes of elastic calls which require b.u., b.u., and b.u., respectively. All calls arrive in the system according to a Poisson process. The call holding time is exponentially distributed with mean value . The initial values of the offered traffic‐load are erl, erl, and erl. Calls of service‐class 3 may retry twice with reduced bandwidth requirement b.u. and b.u., and increased service time so that . In the case of the E‐MRM/BR, we consider the following BR parameters: and so that . We also consider three different values of : (i) b.u., where no bandwidth compression takes place, in which case the E‐MRM and the E‐MRM/BR give exactly the same CBP results with the MRM and the MRM/BR, respectively, (ii) b.u. where bandwidth compression takes place and , and (iii) b.u. where bandwidth compression takes place and . Present graphically the analytical CBP of all service‐classes for the E‐MRM, the E‐MRM/BR, the MRM, and the MRM/BR by assuming that remains constant while increase in steps of 1.0 and 0.5 erl, respectively (up to erl and erl). Also provide simulation CBP results for the E‐MRM and the E‐MRM/BR.
Figures 4.5–4.8 present, for all values of , the analytical and simulation CBP results of service‐classes 1, 2, and 3 (CBP of calls with ), respectively. All figures show that (i) the accuracy of the analytical models is absolutely satisfactory compared to simulation and (ii) the increase of above results in a CBP decrease due to the existence of the compression mechanism.
In the elastic adaptive single‐retry model (EA‐SRM), we consider a link of capacity b.u. that accommodates service‐classes which are distinguished into elastic service‐classes and adaptive service‐classes. Calls of service‐class follow a Poisson process with an arrival rate and have a peak‐bandwidth requirement of b.u. and an exponentially distributed service time with mean . The bandwidth compression/expansion mechanism and the CAC of the EA‐SRM are the same as those of the E‐SRM (Section 4.1.1). The only difference is that adaptive calls do not alter their service time after bandwidth compression/expansion.
Similar to the E‐SRM, the steady state probabilities in the EA‐SRM do not have a PFS, since LB is destroyed between adjacent states (see Figure 4.9). Thus, the unnormalized values of can be determined by an approximate but recursive formula, as presented in Section 4.5.2.
Consider again Example 4.1 (, ) but now service‐class 1 is adaptive. This example clarifies the differences between the E‐SRM and the EA‐SRM.
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with 0.1706 in the E‐SRM) |
The of service‐class 2 calls, , when they require b.u. upon arrival, refers to the percentage of calls which retry and is given by:
(compare with 0.7821 in the E‐SRM) |
The CBP of service‐class 2, , refers to service‐class 2 retry calls which require b.u. and is given by:
(compare with 0.3325 in the E‐SRM) |
The link utilization is determined by:
b.u. (compare with 2.832 in the E‐SRM) |
The comparison between the CBP obtained in the E‐SRM and the EA‐SRM shows that the E‐SRM cannot approximate the EA‐SRM. Furthermore, the CBP of the EA‐SRM are lower, a fact that it is expected since, in this example, adaptive calls of service‐class 1 remain in the system for less time than the corresponding elastic calls.
To facilitate the recursive calculation of , we replace by the state‐dependent compression factors per service‐class and which have already been described in the E‐SRM. The only difference compared to the E‐SRM has to do with the determination of which is now given by [5]:
where and .
Consider again Example 4.13 (, calls of service‐class 1 are adaptive):
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with the exact 0.15286) |
(compare with the exact 0.77806) |
(compare with the exact 0.31627) |
The link utilization is determined by:
b.u. (compare with the exact 2.777) |
Table 4.7 The values of the state‐dependent compression factors and (Example 4.14).
0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 2 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 2 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 2 | 0 | 2 | 1.40 | 0.8571 | 0.00 | 0.7143 |
0 | 0 | 3 | 1.5 | 0.00 | 0.00 | 0.6667 | 2 | 1 | 0 | 1.15 | 0.8696 | 0.8696 | 0.00 |
0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 3 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 1 | 1 | 1.25 | 0.00 | 0.80 | 0.80 | 3 | 0 | 1 | 1.10 | 0.9090 | 0.00 | 0.9090 |
1 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 3 | 1 | 0 | 1.3250 | 0.8679 | 0.7547 | 0.00 |
1 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 4 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 2 | 1.20 | 0.8333 | 0.00 | 0.8333 | 4 | 0 | 1 | 1.2333 | 0.8919 | 0.00 | 0.8108 |
1 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 5 | 0 | 0 | 1.00 | 1.00 | 0.00 | 0.00 |
1 | 1 | 1 | 1.4583 | 0.8571 | 0.6857 | 0.6857 | 6 | 0 | 0 | 1.00 | 1.00 | 0.00 | 0.00 |
To describe the analytical model in the steady state, we consider a link of capacity b.u. that accommodates calls of two service‐classes with traffic parameters: () for service‐class 1 and () for service‐class 2. Service‐class 1 is adaptive while service‐class 2 is elastic. Only calls of service‐class 2 have retry parameters with and . Let be the limit up to which bandwidth compression is permitted for calls of both service‐classes.
Although the EA‐SRM is a non‐PFS model we will use the LB of ( 4.3), initially for calls of service‐class 1. As far as is concerned it is determined by ( 4.4). Based on ( 4.4) and multiplying both sides of ( 4.3) with and , we have:
where and the values of are given by ( 4.43).
Based on the CAC of the EA‐SRM, we consider the following LB equations for calls of service‐class 2:
Equations (4.44), ( 4.8) and ( 4.11) lead to the following system of equations:
Equations (4.45)–(4.47) can be combined into one equation by assuming that calls with are negligible when and calls with are negligible when :
where for , otherwise and for , otherwise .
At this point, we derive a formula for (which is a simplified version of ( 4.43)) by making the following assumptions:
By substituting (4.50) in (4.49) we obtain:
where and are given by (3.8), and by ( 4.1).
Equation (4.51), due to (3.8) and ( 4.1), is written as:
Based on (4.52), we consider again (4.48). Since , when , we have ( 4.16).
When and based on ( 4.52), ( 4.48) can be written as:
since , when .
To introduce the link occupancy distribution in (4.53), we sum both sides of ( 4.53) over the set of states :
Since by definition , (4.54) is written as:
where for .
The combination of ( 4.16) and (4.55) gives the following approximate recursive formula for the calculation of in the case of two service‐classes when service‐class 1 is adaptive, service‐class 2 is elastic, and only calls of service‐class 2 have retry parameters:
where , and for , otherwise , while for , otherwise .
In the case of service‐classes and assuming that all service‐classes may have retry parameters, (4.56) takes the general form [ 5]:
where .
Having determined the unnormalized values of , we can calculate:
Consider again Example 4.13 (, calls of service‐class 1 are adaptive):
The normalization constant is:
. |
The state probabilities are:
(compare with the exact 0.15286) |
(compare with the exact 0.77806) |
(compare with the exact 0.31627) |
(compare with the exact 0.4065(!)) |
It is apparent that even in small EA‐SRM examples the error introduced by the assumption of LB, the introduction of and , and the migration approximation is not significant.
We now consider the EA‐SRM under the BR policy (EA‐SRM/BR) with BR parameter for service‐class calls (). The CAC in the EA‐SRM/BR is the same as that of the E‐SRM/BR. As far as the values of , , and are concerned they are determined by (3.8), ( 4.1), and ( 4.43), respectively.
Consider again Example 4.13 (, calls of service‐class 1 are adaptive) and apply the BR parameters b.u. and b.u. to calls of service‐class 1 and 2, respectively.
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with 0.30737 in the E‐SRM/BR) |
The of service‐class 2 calls, , is given by:
(compare with 0.77264 in the E‐SRM/BR) |
Due to the selection of the BR parameters, the CBP of service‐class 2, , equals :
(compare with 0.30737 in the E‐SRM/BR) |
The link utilization is determined by:
b.u. (compare with 2.771 in the E‐SRM/BR) |
The solution of this linear system is:
Based on the values of , we determine the values of :
Based on the values of , we obtain the following CBP:
(compare with the exact 0.29696) |
(compare with the exact 0.77095) |
(compare with the exact 0.29696) |
The link utilization is determined by:
b.u. (compare with the exact 2.757) |
In the EA‐SRM/BR, the recursive calculation of is based on the Roberts method (see Section 1.3.2.2), which leads to the formula [ 4]:
where .
Based on (4.60), the following performance measures can be calculated:
Consider again Example 4.16 (, calls of service‐class 1 are adaptive).
The normalization constant is:
. |
The state probabilities are:
(compare with the exact 0.29696) |
(compare with the exact 0.77095) |
(compare with the exact 0.29696) |
(compare with the exact 0.38519) |
It is apparent that even in small EA‐SRM/BR examples, the error introduced by the assumption of LB, the introduction of and , the migration approximation, and the application of the BR policy remains acceptable.
Similar to the E‐MRM, in the elastic adaptive multi‐retry model (EA‐MRM) a blocked call of service‐class can have more than one retry parameter for , where and . The call admission in the EA‐MRM is the same as the E‐MRM with the exception that adaptive calls do not alter their service time when their bandwidth is compressed/expanded.
Similar to the EA‐SRM, the steady state probabilities in the EA‐MRM do not have a PFS. Thus, the unnormalized values of can be determined by an approximate but recursive formula, as presented in Section 4.7.2.
Consider again Example 4.6 (, ), and let calls of service‐class 1 be adaptive.
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with 0.470996 in the E‐MRM) |
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
(compare with 0.971428 in the E‐MRM) |
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
(compare with 0.893645 in the E‐MRM) |
The CBP of service‐class 2, , refers to service‐class 2 retry calls which require b.u. and is given by:
(compare with 0.470996 in the E‐MRM) |
The link utilization is determined by:
b.u. (compare with 2.616 in the E‐MRM) |
To facilitate the recursive calculation of , we replace by the state‐dependent compression factors per service‐class and , whose values are given by (3.8) and ( 4.33), respectively. Due to the existence of adaptive traffic, the values of are given by the following formula:
where , and .
Consider again Example 4.18 (, , service‐class 1 is adaptive):
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with the exact 0.456051) |
(compare with the exact 0.9719) |
(compare with the exact 0.894579) |
(compare with the exact 0.456051) |
The link utilization is determined by:
b.u. (compare with the exact 2.617) |
Table 4.8 The values of the state‐dependent compression factors and (Example 4.19).
0 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 3 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 4 | 1.33333 | 0.00 | 0.00 | 0.00 | 0.75 |
0 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 2 | 1.33333 | 0.00 | 0.00 | 0.75 | 0.75 |
0 | 1 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 1 | 0 | 1 | 1.33333 | 0.00 | 0.75 | 0.00 | 0.75 |
1 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 3 | 1.25 | 0.80 | 0.00 | 0.00 | 0.80 |
1 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 1 | 1 | 1.25 | 0.80 | 0.00 | 0.80 | 0.80 |
1 | 1 | 0 | 0 | 1.25 | 0.80 | 0.80 | 0.00 | 0.00 |
2 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 2 | 1.16667 | 0.8571 | 0.00 | 0.00 | 0.8571 |
2 | 0 | 1 | 0 | 1.16667 | 0.8571 | 0.00 | 0.8571 | 0.00 |
3 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
3 | 0 | 0 | 1 | 1.08333 | 0.9231 | 0.00 | 0.00 | 0.9231 |
4 | 0 | 0 | 0 | 1.00 | 1.00 | 0.00 | 0.00 | 0.00 |
Following the analysis of Section 4.5.2.1, the calculation of the unnormalized values of is based on an approximate but recursive formula whose proof is similar to that of ( 4.57) [ 5]:
where , and .
Having determined the unnormalized values of via (4.62) we can calculate [ 5]:
Consider again Example 4.18 (, , service‐class 1 is adaptive):
The normalization constant is:
. |
The state probabilities are:
(compare with the exact 0.456051) |
(compare with the exact 0.9719) |
(compare with the exact 0.894579) |
(compare with the exact 0.456051) |
(compare with the exact 0.46924) |
It is apparent that even in small EA‐MRM examples the error introduced by the assumption of LB, the introduction of and , and the migration approximation is not significant.
Compared to the EA‐SRM/BR, in the elastic adaptive multi‐retry model under the BR policy (EA‐MRM/BR) with BR parameter for service‐class calls (), blocked calls of service‐class can retry more than once to be connected in the system.
Consider again Example 4.18 (, , service‐class 1 is adaptive) and let the BR parameters and so that and .
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with the exact 0.882627 in the E‐MRM/BR) |
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
The of service‐class 2 calls, , when they require b.u. upon arrival, is given by:
(compare with the exact 0.882627 in the E‐MRM/BR) |
The CBP of service‐class 2, , refers to service‐class 2 retry calls which require b.u. and is given by:
(compare with the exact 0.385833 in the E‐MRM/BR) |
The link utilization is determined by:
b.u. (the same as the E‐MRM/BR). |
(compare with the exact 0.39813 in the E‐MRM/BR) |
To facilitate the recursive calculation of in the EA‐MRM/BR, we replace by the state‐dependent compression factors per service‐class and , . The values of and are given by (3.8) and ( 4.33), respectively.
Consider again Example 4.21 (, calls of service‐class 1 are adaptive):
The solution of this linear system is:
Based on the values of , we determine the values of :
(compare with the exact 0.882756) |
(compare with the exact 0.9692) |
(compare with the exact 0.8828) |
(compare with the exact 0.384614) |
The link utilization is determined by:
b.u. (same as the exact) |
(compare with the exact 0.39684) |
Table 4.9 The values of the state‐dependent compression factors and (Example 4.21).
0 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 3 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 0 | 4 | 1.33333 | 0.00 | 0.00 | 0.00 | 0.75 |
0 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 0 | 1 | 2 | 1.33333 | 0.00 | 0.00 | 0.75 | 0.75 |
0 | 1 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
0 | 1 | 0 | 1 | 1.33333 | 0.00 | 0.75 | 0.00 | 0.75 |
1 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 2 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 0 | 3 | 1.25 | 0.80 | 0.00 | 0.00 | 0.80 |
1 | 0 | 1 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
1 | 0 | 1 | 1 | 1.25 | 0.80 | 0.00 | 0.80 | 0.80 |
2 | 0 | 0 | 0 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 1 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
2 | 0 | 0 | 2 | 1.16667 | 0.8571 | 0.00 | 0.00 | 0.8571 |
Following the analysis of Section 4.6.2.1, the calculation of the unnormalized values of is based on an approximate but recursive formula whose proof is similar to that of ( 4.60) [3]:
where
Having determined the unnormalized values of via (4.64) we can calculate:
Consider again Example 4.21 (, , calls of service‐class 1 are adaptive):
The normalization constant is: . The state probabilities are:
(compare with the exact 0.882756) |
(compare with the exact 0.9692) |
(compare with the exact 0.8828) |
(compare with the exact 0.384614) |
(compare with the exact 0.39684) |
It is apparent that in small EA‐MRM/BR examples the error introduced by the assumption of LB, the introduction of and , the migration approximation, and the application of the BR policy can be acceptable (with the exception of and in our example).
Consider again Example 4.12 () and assume that service‐classes 1, 2 are adaptive while service‐class 3 is elastic. Present graphically the analytical CBP of all service‐classes for the EA‐MRM, the EA‐MRM/BR, the MRM, and the MRM/BR by assuming that remains constant while increase in steps of 1.0 and 0.5 erl, respectively (up to erl and erl). In addition, present graphically the link utilization for the EA‐MRM/BR and the MRM/BR. Finally, provide simulation CBP and link utilization results for the EA‐MRM/BR.
Figures 4.11–4.13 present, for all values of , the analytical and simulation CBP results of service‐classes 1, 2, and 3 (CBP of calls with ), respectively, while Figure 4.14 presents the analytical and simulation results for the link utilization. All figures show that the analytical results obtained by the EA‐MRM/BR are of absolutely satisfactory accuracy, compared to simulation, and that the MRM/BR fails to approximate the behavior of the EA‐MRM/BR. This is expected since in the MRM/BR the bandwidth compression/expansion mechanism is not incorporated. Similarly, the results obtained by the MRM and the EA‐MRM fail to approximate the behavior of the EA‐MRM/BR since the BR policy is not applied in these models. Furthermore, Figures 4.11–4.13 show that the existence of the bandwidth compression/expansion mechanism in the EA‐MRM/BR reduces CBP even for small values of . This decrease results in the increase of link utilization in the EA‐MRM/BR compared to the MRM/BR (Figure 4.14).
Since the multirate elastic adaptive retry loss models are a combination of the retry loss models (see Chapter 2) and the elastic adaptive loss models (see Chapter ), the interested reader may refer to Sections 2.11 and 3.7 for possible applications.
Similar to the previous section, the interested reader may refer to the corresponding sections of Chapter 2(Section 2.12) and Chapter 3(Section 3.8). In addition to these sections, interesting extensions of the models presented in this chapter have been proposed in [6]. More precisely, in [ 6] the case of finite sources is considered as well as the application of the BR and TH policies.