Mean value analysis of mixed, multiple class BCMP networks with load dependent service stations

doi:10.1016/0166-5316(84)90010-5

Performance Evaluation

Volume 4, Issue 4, November 1984, Pages 241-260

https://doi.org/10.1016/0166-5316(84)90010-5 Get rights and content

Abstract

In this paper we describe a mean value analysis algorithm to solve product form queueing network models of the Baskett, Chandy, Muntz, and Palacios type (BCMP) [5]. These networks can accommodate multiple job classes and load dependent service stations. The networks can be closed (i.e., jobs neither enter nor leave the network), open (i.e., jobs can enter and leave the network), or mixed (i.e., closed with respect to some job classes and open with respect to others). Two major aspects of this paper are:

1.
(1)to present a unified approach to the solution of this class of BCMP queueing networks [5], and
2.
(2)to thereby extend and consolidate the results of the work of several authors (Balbo, Bruell and Schwetman (1977), Bruell and Balbo (1980), Chandy and Sauer (1980), Krzesinki, Teunissen and Kritzinger (1981), Resier and Kobayashi (1975), Reiser and Lavenberg (1980), Reiser (1981), Sauer (1983), Schwetman (1980), Tucci and MacNair (1982) and Zahorjan and Wong (1981)).

The paper concludes by collecting together all the results into an explicit algorithm, MVALDMX (for mean value analysis load dependent mixed) that is a generalization of the original MVA algorithm of Reiser and Lavenberg (1978, 1980).

References (32)

M. Reiser
Mean value analysis and convolution method for queueing dependent servers in closed queueing networks
Performance Evaluation
(1981)
P.V. Afshari et al.
On modeling a new bus accessing scheme, Comput.
P.V. Afshari et al.
The load balancing bus accessing scheme
IEEE Trans. Comput.
(1983)
G. Balbo et al.
Customer classes and closed network models: A solution technique
G. Balbo, S.C. Bruell, L. Cerchio, D. Chiaberto and L. Molinatti, Mean value analysis of closed load dependent queueing...
F. Baskett et al.
Open, closed and mixed networks of queues with different classes of customers
J. ACM
(1975)
S.C. Bruell
On single and multiple job class queueing network models of computer systems
S.C. Bruell et al.
Computational algorithms for closed queueing networks
J.P. Buzen
Queueing network models of multiprogramming
J.P. Buzen
Computational algorithms for closed queueing networks with exponential servers
Comm. ACM
(1973)

K.M. Chandy et al.

Computational algorithms for product form queueing networks

Comm. ACM

(1980)

J.P. Buzen et al.

Operational treatment of queue distributions and mean value analysis

Comput. Performance

(1980)

W.J. Gordon et al.

Closed queueing systems with exponential servers

Oper. Res.

(1967)

G.S. Graham

K.P. Hoyme, S.C. Bruell, P.V. Afshari and R.Y. Kain, A tree-structured MVA algorithm, University of Minnesota,...

J.R. Jackson

Jobshop-like queueing systems

Oper. Res.

(1963)

Cited by (33)

Facilitating load-dependent queueing analysis through factorization
2021, Performance Evaluation
Citation Excerpt :
Unfortunately, despite a long history of work on this subject [12], when LLD stations are considered in multiclass closed networks, which are used to abstract systems with finite levels of parallelism, they become expensive to analyze. Indeed, commonly employed evaluation techniques for models with fixed-rate stations, such as the convolution algorithm (CA) [13–15] and mean value analysis (MVA) [16] or its approximate versions [17], are much more expensive to apply when extended to the LLD setting. These methods can also incur numerical instabilities [18], as we further discuss in Section 8.
We propose novel exact and approximate solutions for mean-value and probabilistic analysis of closed queueing networks with limited load-dependent nodes. The main result is the derivation of an exact correction factor between the equilibrium solution of a load-dependent model and the one of a related model with fixed-rate queueing stations. This enables the reuse of state-of-the-art methods for fixed-rate stations in the computation of performance metrics for load-dependent systems. As many such algorithms are available, our findings significantly increase the range of techniques available to study load-dependence. We further interpret the correction factor as a load-dependent normalizing constant and propose two novel integral forms, in the real and in the complex domains, which can be used for its efficient computation. These integral forms are also shown to be applicable to evaluate more general limited load-dependent systems, as we illustrate in a sojourn time distribution analysis problem. Lastly, the proposed algorithms are numerically examined through thousands of experiments, which reveal accuracy in the range 1%–6% mean absolute relative error.
Power control in saturated fork-join queueing systems
2017, Performance Evaluation
The analysis of fork-join queueing systems has played an important role for the performance evaluation of distributed systems where parallel computations associated with the same job are carried out and a job is considered served only when all the parallel tasks it consists of are served and then joined. The fork-join nodes that we consider consist of $K \geq 2$ parallel servers each of which is equipped with two First Come First Served queues, namely the service-queue and the join-queue. The former stores the tasks waiting to be served while the latter stores the served tasks waiting to be joined. Under heavy load conditions, the variance of the service times associated with the tasks tends to cause long join-queue lengths. In this work, we propose an algorithm to dynamically control the servers’ speeds (e.g., via frequency scaling), that aims at reducing the power consumption of the servers whose join-queue lengths are longer than the others’. Under Markovian assumptions, we provide a model for the performance evaluation of the system in saturation that allows us to derive the expression for the steady-state distribution, the system’s throughput and balance index. Finally, we derive the analytical expression for the marginal state probabilities of each server and provide upper and lower bounds for the expected power consumption.
QD-AMVA: Evaluating systems with queue-dependent service requirements
2015, Performance Evaluation
Citation Excerpt :
To support the raising interest for characterizing load dependence in queueing networks, this paper introduces QD-AMVA, a novel AMVA algorithm for queue-dependent models, i.e., models of systems where the service requirements depend on the number and mix of requests in execution at the resource where the request executes. Although AMVA methods for load-dependent models exist [1], these methods have severe limitations in the multiclass setting, including excessive computational requirements, numerical instabilities, and being restricted to the simplest types of queue-dependence [16,17]. This has led to multiclass state-dependent networks being seldom used in real-world studies.
Workload measurements in enterprise systems often lead to observe a dependence between the number of requests running at a resource and their mean service requirements. However, multiclass performance models that feature these dependences are challenging to analyze, a fact that discourages practitioners from characterizing workload dependences. We here focus on closed multiclass queueing networks and introduce QD-AMVA, the first approximate mean-value analysis (AMVA) algorithm that can efficiently and robustly analyze queue-dependent service times in a multiclass setting. A key feature of QD-AMVA is that it operates on mean values, avoiding the computation of state probabilities. This property is an innovative result for state-dependent models, which increases the computational efficiency and numerical robustness of their evaluation. Extensive validation on random examples, a cloud load-balancing case study and comparison with a fluid method and an existing AMVA approximation prove that QD-AMVA is efficient, robust and easy to apply, thus enhancing the tractability of queue-dependent models.
Time-bounds for production networks
2006, International Journal of Production Economics
Lower and upper time bounds are established for completing a task using a system that can be structurally represented by a digraph in which the nodes are production components. Each component is assumed to execute a known number of operations and each operation is assumed to require a known delay. The algorithms presented provide a rapid means for computing an objective function when optimizing a system using a method that requires the objective function to be evaluated an extremely large number of times, such as a simulated annealing or genetic algorithm.
An approximate MVA algorithm for exponential, class-dependent multiple servers
1997, Performance Evaluation
Mean value analysis (MVA) is an efficient algorithm for determining the mean sojourn time, the mean queue length, and the throughput in a closed multiclass queueing network. It provides exact results for the class of product-form networks. Often different classes have different service requirements in FCFS queues, but such networks are not of product form. There are several possibilities to compute performance measure for such nodes and networks. In this paper we present an approximation formula for multiple-server FCFS queues with class-dependent service times as a Norton flow equivalent product node, where the departure rate of any class depends on the number of customers of all classes in the queue. We will use this approximation in the sojourn time formula of some exact and approximate MVA algorithms.
MVA approximations for SJN scheduling
1992, Performance Evaluation
The paper proposes an approximation for modeling shortest-job-next (SJN) scheduling discipline at single server stations. The approximation is developed in the form of a response time equation, which can be used directly in the well known Mean Value Analysis (MVA) technique. Detailed experimental results are presented that show the approximation to be highly accurate. We then examine a more complex situation, i.e., the case of a multiple chain network with nonpreemptive priority among chains and SJN scheduling within a chain. Two known ways of modeling priorities are considered and the results compared. Finally, we propose two techniques for handling stations with a restricted type of load dependence. These techniques are not limited to SJN scheduling, but apply to any discipline for which we can write reliable response time equations for a load-independent station.

View all citing articles on Scopus

^☆: This work was supported in part by NSF Grant MCS-80006308 and Nato Research Grant 280.81.

View full text

Mean value analysis of mixed, multiple class BCMP networks with load dependent service stations☆

Abstract

Performance Evaluation

On modeling a new bus accessing scheme, Comput.

The load balancing bus accessing scheme

IEEE Trans. Comput.

Customer classes and closed network models: A solution technique

Open, closed and mixed networks of queues with different classes of customers

J. ACM

On single and multiple job class queueing network models of computer systems

Computational algorithms for closed queueing networks

Queueing network models of multiprogramming

Computational algorithms for closed queueing networks with exponential servers

Comm. ACM

Computational algorithms for product form queueing networks

Comm. ACM

Operational treatment of queue distributions and mean value analysis

Comput. Performance

Closed queueing systems with exponential servers

Oper. Res.

Jobshop-like queueing systems

Oper. Res.