-
On Optimal Server Allocation for Moldable Jobs with Concave Speed-Up
Authors:
Samira Ghanbarian,
Arpan Mukhopadhyay,
Ravi R. Mazumdar,
Fabrice M. Guillemin
Abstract:
A large proportion of jobs submitted to modern computing clusters and data centers are parallelizable and capable of running on a flexible number of computing cores or servers. Although allocating more servers to such a job results in a higher speed-up in the job's execution, it reduces the number of servers available to other jobs, which in the worst case, can result in an incoming job not findin…
▽ More
A large proportion of jobs submitted to modern computing clusters and data centers are parallelizable and capable of running on a flexible number of computing cores or servers. Although allocating more servers to such a job results in a higher speed-up in the job's execution, it reduces the number of servers available to other jobs, which in the worst case, can result in an incoming job not finding any available server to run immediately upon arrival. Hence, a key question to address is: how to optimally allocate servers to jobs such that (i) the average execution time across jobs is minimized and (ii) almost all jobs find at least one server immediately upon arrival. To address this question, we consider a system with $n$ servers, where jobs are parallelizable up to $d^{(n)}$ servers and the speed-up function of jobs is concave and increasing. Jobs not finding any available servers upon entry are blocked and lost. We propose a simple server allocation scheme that achieves the minimum average execution time of accepted jobs while ensuring that the blocking probability of jobs vanishes as the system becomes large ($n \to \infty$). This result is established for various traffic conditions as well as for heterogeneous workloads. To prove our result, we employ Stein's method which also yields non-asymptotic bounds on the blocking probability and the mean execution time. Furthermore, our simulations show that the performance of the scheme is insensitive to the distribution of job execution times.
△ Less
Submitted 15 April, 2024;
originally announced June 2024.
-
On the Performance of Large Loss Systems with Adaptive Multiserver Jobs
Authors:
Samira Ghanbarian,
Arpan Mukhopadhyay,
Fabrice M. Guillemin,
Ravi R. Mazumdar
Abstract:
In this paper, we study systems where each job or request can be split into a flexible number of sub-jobs up to a maximum limit. The number of sub-jobs a job is split into depends on the number of available servers found upon its arrival. All sub-jobs of a job are then processed in parallel at different servers leading to a linear speed-up of the job. We refer to such jobs as {\em adaptive multi-s…
▽ More
In this paper, we study systems where each job or request can be split into a flexible number of sub-jobs up to a maximum limit. The number of sub-jobs a job is split into depends on the number of available servers found upon its arrival. All sub-jobs of a job are then processed in parallel at different servers leading to a linear speed-up of the job. We refer to such jobs as {\em adaptive multi-server jobs}. We study the problem of optimal assignment of such jobs when each server can process at most one sub-job at any given instant and there is no waiting room in the system.
We assume that, upon arrival, a job can only access a randomly sampled subset of $k(n)$ servers from a total of $n$ servers, and the number of sub-jobs is determined based on the number of idle servers within the sampled subset. We analyze the steady-state performance of the system when system load varies according to $λ(n) =1 - βn^{-α}$ for $α\in [0,1)$, and $β\geq 0$. Our interest is to find how large the subset $k(n)$ should be in order to have zero blocking and maximum speed-up in the limit as $n \to \infty$. We first characterize the system's performance when the jobs have access to the full system, i.e., $k(n)=n$. In this setting, we show that the blocking probability approaches to zero at the rate $O(1/\sqrt{n})$ and the mean response time of accepted jobs approaches to its minimum achievable value at rate $O(1/n)$. We then consider the case where the jobs only have access to subset of servers, i.e., $k(n) < n$. We show that as long as $k(n)=ω(n^α)$, the same asymptotic performance can be achieved as in the case with full system access. In particular, for $k(n)=Θ(n^α\log n)$, we show that both the blocking probability and the mean response time approach to their desired limits at rate $O(n^{-(1-α)/2})$.
△ Less
Submitted 31 August, 2023;
originally announced September 2023.
-
On the number of departures from the $M/M/\infty$ queue in a finite time interval
Authors:
Fabrice Guillemin
Abstract:
In this paper, we analyze the number of departures from an initially empty $M/M/\infty$ system in a finite time interval. We observe the system during an exponentially distributed period of time starting from the time origin. We then consider the absorbed Markov chain describing the number of arrivals and departures in the system until the observer leaves the system, triggering the absorption of t…
▽ More
In this paper, we analyze the number of departures from an initially empty $M/M/\infty$ system in a finite time interval. We observe the system during an exponentially distributed period of time starting from the time origin. We then consider the absorbed Markov chain describing the number of arrivals and departures in the system until the observer leaves the system, triggering the absorption of the Markov chain. The generator of the absorbed Markov chain induces a selfadjoint operator in some Hilbert space. The use of spectral theory then allows us to compute the Laplace transform of several transient characteristics of the $M/M/\infty$ system (namely, the number of transitions of the Markov chain until absorption, the number of departures from the system, etc.). The analysis is extended to the finite capacity $MM/c/c$ system for some finite integer $c$}.
△ Less
Submitted 17 March, 2023; v1 submitted 10 August, 2022;
originally announced August 2022.
-
Asymptotic analysis of the sojourn time of a batch in an $M^{[X]}/M/1$ Processor Sharing Queue
Authors:
Fabrice Guillemin,
Alain Simonian,
Ridha Nasri,
Veronica Quintuna Rodriguez
Abstract:
In this paper, we exploit results obtained in an earlier study for the Laplace transform of the sojourn time $Ω$ of an entire batch in the $M^{[X]}/M/1$ Processor Sharing (PS) queue in order to derive the asymptotic behavior of the complementary probability distribution function of this random variable, namely the behavior of $P(Ω>x)$ when $x$ tends to infinity. We precisely show that up to a mult…
▽ More
In this paper, we exploit results obtained in an earlier study for the Laplace transform of the sojourn time $Ω$ of an entire batch in the $M^{[X]}/M/1$ Processor Sharing (PS) queue in order to derive the asymptotic behavior of the complementary probability distribution function of this random variable, namely the behavior of $P(Ω>x)$ when $x$ tends to infinity. We precisely show that up to a multiplying factor, the behavior of $P(Ω>x)$ for large $x$ is of the same order of magnitude as $P(ω>x)$, where $ω$ is the sojourn time of an arbitrary job is the system. From a practical point of view, this means that if a system has to be dimensioned to guarantee processing time for jobs then the system can also guarantee processing times for entire batches by introducing a marginal amount of processing capacity.
△ Less
Submitted 19 April, 2021;
originally announced April 2021.
-
Inversion of a Class of Singular Integral Operators on Entire Functions
Authors:
Ridha Nasri,
Alain Simonian,
Fabrice Guillemin
Abstract:
Given constants $x, ν\in \mathbb{C}$ and the space $\mathscr{H}_0$ of entire functions in $\mathbb{C}$ vanishing at $0$, we consider the integro-differential operator $$ \mathfrak{L} = \left ( \frac{x \, ν(1-ν)}{1-x} \right ) \; δ\circ \mathfrak{M}\, , $$ with $δ= z \, \mathrm{d}/\mathrm{d}z$ and $\mathfrak{M}:\mathscr{H}_0 \rightarrow \mathscr{H}_0$ defined by…
▽ More
Given constants $x, ν\in \mathbb{C}$ and the space $\mathscr{H}_0$ of entire functions in $\mathbb{C}$ vanishing at $0$, we consider the integro-differential operator $$ \mathfrak{L} = \left ( \frac{x \, ν(1-ν)}{1-x} \right ) \; δ\circ \mathfrak{M}\, , $$ with $δ= z \, \mathrm{d}/\mathrm{d}z$ and $\mathfrak{M}:\mathscr{H}_0 \rightarrow \mathscr{H}_0$ defined by $$ \mathfrak{M}f(z) = \int_0^1 e^{-z t^{-ν}(1-(1-x)t)} \, f \left (z \, t^{-ν}(1-t) \right ) \, \frac{\mathrm{d}t}{t}, \qquad z \in \mathbb{C}, $$ for any $f \in \mathscr{H}_0$. Operator $\mathfrak{L}$ originates from an inversion problem in Queuing Theory. Bringing the inversion of $\mathfrak{L}$ back to that of $\mathfrak{M}$ translates into a singular Volterra integral equation, but with no explicit kernel.
In this paper, the inverse of operator $\mathfrak{L}$ is derived through a new inversion formula recently obtained for infinite matrices with entries involving Hypergeometric polynomials. For $x \notin \mathbb{R}^- \cup \{1\}$ and $\mathrm{Re}(ν) < 0$, we then show that the inverse $\mathfrak{L}^{-1}$ of $\mathfrak{L}$ on $\mathscr{H}_0$ has the integral representation $$ \mathfrak{L}^{-1}g(z) = \frac{1-x}{2iπx} \, e^{z} \int_1^{(0+)} \frac{e^{-xtz}}{t(t-1)} \, g \left (z \, (-t)^ν(1-t)^{1-ν} \right ) \, \mathrm{d}t, \qquad z \in \mathbb{C}, $$ for any $g \in \mathscr{H}_0$, where the bounded integration contour in the complex plane starts at point 1 and encircles the point 0 in the positive sense. Other related integral representations of $\mathfrak{L}^{-1}$ are also provided.
△ Less
Submitted 4 January, 2021;
originally announced January 2021.
-
On the sojourn time of a batch in the $M^{[X]}/M/1$ Processor Sharing Queue
Authors:
Fabrice Guillemin,
Alain Simonian,
Ridha Nasri,
Veronica Quintuna Rodriguez
Abstract:
In this paper, we analyze the sojourn of an entire batch in a processor sharing $M^{[X]}/M/1$ processor queue, where geometrically distributed batches arrive according to a Poisson process and jobs require exponential service times. By conditioning on the number of jobs in the systems and the number of jobs in a tagged batch, we establish recurrence relations between conditional sojourn times, whi…
▽ More
In this paper, we analyze the sojourn of an entire batch in a processor sharing $M^{[X]}/M/1$ processor queue, where geometrically distributed batches arrive according to a Poisson process and jobs require exponential service times. By conditioning on the number of jobs in the systems and the number of jobs in a tagged batch, we establish recurrence relations between conditional sojourn times, which subsequently allow us to derive a partial differential equation for an associated bivariate generating function. This equation involves an unknown generating function, whose coefficients can be computed by solving an infinite lower triangular linear system. Once this unknown function is determined, we compute the Laplace transform and the mean value of the sojourn time of a batch in the system.
△ Less
Submitted 28 September, 2020;
originally announced September 2020.
-
A New Linear Inversion Formula for a class of Hypergeometric polynomials
Authors:
Ridha Nasri,
Alain Simonian,
Fabrice Guillemin
Abstract:
Given complex parameters $x$, $ν$, $α$, $β$ and $γ\notin -\mathbb{N}$, consider the infinite lower triangular matrix $\mathbf{A}(x,ν;α, β,γ)$ with elements $$ A_{n,k}(x,ν;α,β,γ) = \displaystyle (-1)^k\binom{n+α}{k+α} \cdot F(k-n,-(β+n)ν;-(γ+n);x) $$ for $1 \leqslant k \leqslant n$, depending on the Hypergeometric polynomials $F(-n,\cdot;\cdot;x)$, $n \in \mathbb{N}^*$. After stating a general crit…
▽ More
Given complex parameters $x$, $ν$, $α$, $β$ and $γ\notin -\mathbb{N}$, consider the infinite lower triangular matrix $\mathbf{A}(x,ν;α, β,γ)$ with elements $$ A_{n,k}(x,ν;α,β,γ) = \displaystyle (-1)^k\binom{n+α}{k+α} \cdot F(k-n,-(β+n)ν;-(γ+n);x) $$ for $1 \leqslant k \leqslant n$, depending on the Hypergeometric polynomials $F(-n,\cdot;\cdot;x)$, $n \in \mathbb{N}^*$. After stating a general criterion for the inversion of infinite matrices in terms of associated generating functions, we prove that the inverse matrix $\mathbf{B}(x,ν;α, β,γ) = \mathbf{A}(x,ν;α, β,γ)^{-1}$ is given by \begin{align} B_{n,k}(x,ν;α, β,γ) = & \; \displaystyle (-1)^k\binom{n+α}{k+α} \; \cdot \nonumber \\ & \; \biggl [ \; \frac{γ+k}{β+k} \, F(k-n,(β+k)ν;γ+k;x) \; + \nonumber \\ & \; \; \; \frac{β-γ}{β+k} \, F(k-n,(β+k)ν;1+γ+k;x) \; \biggr ] \nonumber \end{align} for $1 \leqslant k \leqslant n$, thus providing a new class of linear inversion formulas. Functional relations for the generating functions of related sequences $S$ and $T$, that is, $T = \mathbf{A}(x,ν;α, β,γ) \, S \Longleftrightarrow S = \mathbf{B}(x,ν;α, β,γ) \, T$, are also provided.
△ Less
Submitted 5 July, 2020;
originally announced July 2020.
-
Sojourn time in a $M^{[X]}/M/1$ Processor Sharing Queue with batch arrivals (II)
Authors:
F. Guillemin,
V. K. Quintuna Rodriguez,
A. Simonian,
R. Nasri
Abstract:
For the $M^{[X]}/M/1$ processor Sharing queue with batch arrivals, the sojourn time $Ω$ of a batch is investigated.
We first show that the distribution of $Ω$ can be generally obtained from an infinite linear differential system. When further assuming that the batch size has a geometric distribution with given parameter $q \in [0,1[$, this differential system is further analyzed by means of an a…
▽ More
For the $M^{[X]}/M/1$ processor Sharing queue with batch arrivals, the sojourn time $Ω$ of a batch is investigated.
We first show that the distribution of $Ω$ can be generally obtained from an infinite linear differential system. When further assuming that the batch size has a geometric distribution with given parameter $q \in [0,1[$, this differential system is further analyzed by means of an associated bivariate generating function $(x,u,v) \mapsto E(x,u,v)$. Specifically, denoting by $s \mapsto E^*(s,u,v)$ the one-sided Laplace transform of $E(\cdot,u,v)$ and defining $$ Φ(s,u,v) = P(s,u) \, (1-v) \, F^*(s,u,uv), \quad 0 < \vert u \vert < 1, \, \vert v \vert < 1, $$ for some known polynomial $P(s,u)$ and where $$ F^*(s,u,v) = \frac{E^*(s,u,v)-E^*(s,q,v)}{u-q}, $$ we show that the function $Φ$ verifies an inhomogeneous linear partial differential equation (PDE) $$ \frac{\partial Φ}{\partial u} - \left [ \frac{u - q}{P(s,u)} \right ] v(1-v) \, \frac{\partial Φ}{\partial v} + \ell(s,u,v) = 0 $$ for given $s$, where the last term $\ell(s,u,v)$ involves both $E^*(s,q,v)$ and the first order derivative $\partial E^*(s,q,v)/\partial v$ at the boundary point $u = q$. Solving this PDE for $Φ$ via its characteristic curves and with the required analyticity properties eventually determines the one-sided Laplace transform $E^*$.
By means of a Laplace inversion of this transform $E^*$, the distribution function of the sojourn time $Ω$ of a batch is then given in an integral form. The tail behavior of the distribution of sojourn time $Ω$ is finally derived.
△ Less
Submitted 3 June, 2020;
originally announced June 2020.
-
An inversion formula with hypergeometric polynomials and application to singular integral operators
Authors:
R. Nasri,
A. Simonian,
F. Guillemin
Abstract:
Given parameters $x \notin \mathbb{R}^- \cup \{1\}$ and $ν$, $\mathrm{Re}(ν) < 0$, and the space $\mathscr{H}_0$ of entire functions in $\mathbb{C}$ vanishing at $0$, we consider the family of operators $\mathfrak{L} = c_0 \cdot δ\circ \mathfrak{M}$ with constant $c_0 = ν(1-ν)x/(1-x)$, $δ= z \, \mathrm{d}/\mathrm{d}z$ and integral operator $\mathfrak{M}$ defined by…
▽ More
Given parameters $x \notin \mathbb{R}^- \cup \{1\}$ and $ν$, $\mathrm{Re}(ν) < 0$, and the space $\mathscr{H}_0$ of entire functions in $\mathbb{C}$ vanishing at $0$, we consider the family of operators $\mathfrak{L} = c_0 \cdot δ\circ \mathfrak{M}$ with constant $c_0 = ν(1-ν)x/(1-x)$, $δ= z \, \mathrm{d}/\mathrm{d}z$ and integral operator $\mathfrak{M}$ defined by $$ \mathfrak{M}f(z) = \int_0^1 e^{- \frac{z}{x}t^{-ν}(1-(1-x)t)} \, f \left ( \frac{z}{x} \, t^{-ν}(1-t) \right ) \, \frac{\mathrm{d}t}{t}, \qquad z \in \mathbb{C}, $$ for all $f \in \mathscr{H}_0$. Inverting $\mathfrak{L}$ or $\mathfrak{M}$ proves equivalent to solve a singular Volterra equation of the first kind.
The inversion of operator $\mathfrak{L}$ on $\mathscr{H}_0$ leads us to derive a new class of linear inversion formulas $T = A(x,ν) \cdot S \Leftrightarrow S = B(x,ν) \cdot T$ between sequences $S = (S_n)_{n \in \mathbb{N}^*}$ and $T = (T_n)_{n \in \mathbb{N}^*}$, where the infinite lower-triangular matrix $A(x,ν)$ and its inverse $B(x,ν)$ involve Hypergeometric polynomials $F(\cdot)$, namely $$
\left\{
\begin{array}{ll}
A_{n,k}(x,ν) = \displaystyle (-1)^k\binom{n}{k}F(k-n,-nν;-n;x),
B_{n,k}(x,ν) = \displaystyle (-1)^k\binom{n}{k}F(k-n,kν;k;x)
\end{array} \right. $$ for $1 \leqslant k \leqslant n$. Functional relations between the ordinary (resp. exponential) generating functions of the related sequences $S$ and $T$ are also given. These relations finally enable us to derive the integral representation $$ \mathfrak{L}^{-1}f(z) = \frac{1-x}{2iπx} \, e^{z} \int_{(0+)}^1 \frac{e^{-xtz}}{t(1-t)} \, f \left ( xz \, (-t)^ν(1-t)^{1-ν} \right ) \, \mathrm{d}t, \quad z \in \mathbb{C}, $$ for the inverse $\mathfrak{L}^{-1}$ of operator $\mathfrak{L}$ on $\mathscr{H}_0$, where the integration contour encircles the point 0.
△ Less
Submitted 19 September, 2019;
originally announced September 2019.
-
Inversion formula with hypergeometric polynomials and its application to an integral equation
Authors:
Ridha Nasri,
Alain Simonian,
Fabrice Guillemin
Abstract:
For any complex parameters $x$ and $ν$, we provide a new class of linear inversion formulas $T = A(x,ν) \cdot S \Leftrightarrow S = B(x,ν) \cdot T$ between sequences $S = (S_n)_{n \in \mathbb{N}^*}$ and $T = (T_n)_{n \in \mathbb{N}^*}$, where the infinite lower-triangular matrix $A(x,ν)$ and its inverse $B(x,ν)$ involve Hypergeometric polynomials $F(\cdot)$, namely…
▽ More
For any complex parameters $x$ and $ν$, we provide a new class of linear inversion formulas $T = A(x,ν) \cdot S \Leftrightarrow S = B(x,ν) \cdot T$ between sequences $S = (S_n)_{n \in \mathbb{N}^*}$ and $T = (T_n)_{n \in \mathbb{N}^*}$, where the infinite lower-triangular matrix $A(x,ν)$ and its inverse $B(x,ν)$ involve Hypergeometric polynomials $F(\cdot)$, namely $$
\left\{
\begin{array}{ll}
A_{n,k}(x,ν) = \displaystyle (-1)^k\binom{n}{k}F(k-n,-nν;-n;x),
\\
B_{n,k}(x,ν) = \displaystyle (-1)^k\binom{n}{k}F(k-n,kν;k;x)
\end{array} \right. $$ for $1 \leqslant k \leqslant n$. Functional relations between the ordinary (resp. exponential) generating functions of the related sequences $S$ and $T$ are also given.
These new inversion formulas have been initially motivated by the resolution of an integral equation recently appeared in the field of Queuing Theory; we apply them to the full resolution of this integral equation. Finally, matrices involving generalized Laguerre polynomials polynomials are discussed as specific cases of our general inversion scheme.
△ Less
Submitted 16 April, 2019;
originally announced April 2019.
