Jaderick P. Pabico

A novel dynamic load balancing library for cluster computing

In the last few years, research advances in dynamic scheduling at application and runtime system levels have contributed to improving the performance of scientific applications in heterogeneous environments. This paper presents the design and implementation of a library as a result of an integrated approach to dynamic load balancing. This approach combines the advantages of optimizing data migration via novel dynamic loop scheduling strategies with the advances in object migration mechanisms of parallel runtime systems. The performance improvements obtained by the use of this library have been investigated by its use in two scientific applications: the N-body simulations, and the profiling of automatic quadrature routines. The experimental results obtained underscore the significance of using such an integrated approach, as well as the benefits of using the library especially in cluster applications characterized by irregular and unpredictable behavior.

Design and implementation of a novel dynamic load balancing library for cluster computing

This paper presents the design and implementation of a library based on an integrated approach to dynamic load balancing. This approach combines the advantages of optimizing data migration via novel dynamic loop scheduling strategies with the advances in resource management and task migration capabilities offered by a recently developed parallel runtime system. The performance improvements obtained by the use of this library have been investigated by its use in three scientific applications: the N-body simulations, the profiling of automatic quadrature routines, and the heat solver in an unstructured grid. The experimental results obtained underscore the significance of using such an integrated approach, as well as the benefits of using the library especially in applications characterized by irregular and unpredictable behavior.

Overhead analysis of a dynamic load balancing library for cluster computing

This paper investigates the overhead of a dynamic load balancing library for large irregular data-parallel scientific applications on general-purpose clusters. The library is based on an integrated approach combining the advantages of novel dynamic loop scheduling strategies as data migration policies with the advances in resource management and task migration capabilities offered by a recently developed parallel runtime system. The paper focuses on the contribution of the runtime system software layer to the total overhead of the library. Experiments to compare the performance of two applications using the library, the N-body simulations and the profiling of a quadrature routine, with the performance of the same applications using an MPI-only implementation of the dynamic scheduling techniques indicate only a slight decrease in performance due to the overhead of the runtime system software layer. The results validate the suitability of the runtime system as an implementation platform for dynamic load balancing schemes, and underscore the significance of using the integrated approach, as well as the benefits of using the library especially in cluster applications characterized by irregular and unpredictable behavior.

Automatic selection of loop scheduling algorithms using reinforcement learning

This paper presents the design and implementation of a reinforcement learning agent that automatically selects appropriate loop scheduling algorithms for parallel loops embedded in time-stepping scientific applications executing on clusters. There may be a number of such loops in an application, and the loops may have different load balancing requirements. Further, loop characteristics may also change as the application progresses. Following a model-free learning approach, the learning agent assigned to a loop selects from a library the best scheduling algorithm for the loop during the lifetime of the application. The utility of the learning agent is demonstrated by its successful integration into the simulation of wave packets - an application arising from quantum mechanics. Results of statistical analysis using pairwise comparison of means on the running time of the simulation with and without the learning agent validate the effectiveness of the agent in improving the parallel performance of the simulation.

Human-Borg dynamics during a cybernetic alien invasion

We propose a series of ODE systems to model the various dynamics of humanBorg interaction during a Borg invasion. These models, progressively developed one after the other, could provide humanity and other peace-loving intergalactic species the necessary mathematical tools to develop survival strategies in the event of future alien invasion. The Borg is a race of technologically-advanced cybernetic aliens and acts as a very powerful antagonist against the peace-loving human species in various Star Trek sci-fi story lines. Cybernetics, also called cyborgs, are organic individuals implanted with intelligent electromechanical devices for the purpose of increasing the individuals’ efficiency by several degrees (e.g., strength, speed, and intelligence) but in the expense of procreation. Thus, the “parasitic” Borg needs to assimilate other species in an epidemiological manner for the survival of their own race. In these models, humans can be transformed into one of six types depending on their reaction on or resistance to Borg assimilation. These are Susceptible (S), Captured (C), Assimilated (A), Rescued (R), Educated (or Rehabilitated, E), and Defiant (D). Susceptible humans can be captured and then assimilated into being a Borg drone. The remaining humans can rescue those who were captured or assimilated. Once rescued, they will undergo rehabilitation after which they either end up (again) susceptible to or strongly defiant from being captured and assimilated. We start by describing the SCA model which has the same (analytical and/or numerical) solution to the Susceptible-Exposed-Infected model in epidemiology. Then we move on to the SCAR model which incorporates the tendency of humans to fight back by rescuing the captured or assimilated. SCARE further models the propensity of humans to educate (or rehabilitate) those whom they have rescued. Finally, we present the SCARED model which describes the natural inclinations of humans to either “relapse” to being susceptible to or grow being defiant against assimilation after undergoing rehabilitation. The numerical solutions to all these models will be presented using a popular yet simple computer software. The SCA, SCAR, and SCARE models are reduction from the SCARED model when all the respective coefficients of the quantities not present in the reduced model are zero. The bottomline is SCA ⊂ SCAR ⊂ SCARE ⊂ SCARED. The dynamics in the general SCARED model is governed by the system of ODEs shown (as a teaser) in the attached addendum. ∗Submitted as a scientific oral paper contribution to the 2017 MSP CALABARZON Annual Regional Convention, First Asia Institute of Technology and Humanities, Tanauan City, Batangas, 23 September 2017. 1 PeerJ Preprints | https://doi.org/10.7287/peerj.preprints.3198v1 | CC BY 4.0 Open Access | rec: 28 Aug 2017, publ: 28 Aug 2017 Addendum to the Abstract: Human-Borg Dynamics During a Cybernetic Alien Invasion The human-borg dynamics under the SCARED model is governed by the following system of ODEs: d dt S(t) = β {S(t) +D(t)}+ ζE(t)− δ1S(t)− γS(t)A(t) (1) d dt C(t) = γS(t)A(t)− δ2C(t)− ρcC(t) {S(t) + 2D(t)} − αC(t) (2) d dt A(t) = αC(t)− δ3A(t)− ρaA(t) {S(t) + 2D(t)} − δ6A(t)D(t) (3) d dt R(t) = ρaA(t) {S(t) + 2D(t)}+ ρcC(t) {S(t) + 2D(t)} − δ4R(t)− ξR(t) (4) d dt E(t) = ξR(t)− δ5E(t)− ζE(t)− (1− ζ)D(t) (5) d dt D(t) = (1− ζ)D(t)− δ1D(t)− δ7A(t)D(t) (6) The functions S(t), C(t), A(t), R(t), E(t), and D(t) are the respective population counts of susceptible, captured, assimilated, rescued, educated, and defiant humans at any time t ≥ 0. All coefficients are ∈ [0, 1] and are explained in the table below. Coefficient Description β Natural birth rate of S ∪D δ1 Natural death rate of S and D δ2 Increased death rate due to captivity δ3 Decreased death rate due to physical perfection as a result of assimilation δ4 Increased death rate due to rescue operations δ5 Decreased death rate due to extra protection during education δ6 Death rate of A when it comes in contact with a D δ7 Death rate of D when it comes in contact with A α Rate of assimilation of C γ Rate of captivity when S and A come into contact with each other ζ Probability by which E becomes S again ξ Rate of education ρc Rate of rescue of C by either S or D ρa Rate of rescue of A by either S or D 2 PeerJ Preprints | https://doi.org/10.7287/peerj.preprints.3198v1 | CC BY 4.0 Open Access | rec: 28 Aug 2017, publ: 28 Aug 2017

Profiling waitlisted incoming students for future delinquency with an ensemble of statistical machine learning algorithms.

Given a dataset R = {R1, R2, . . . , Rr} of r records of waitlisted incoming freshman students (WIFS), where for any i = 1, 2, . . . , r, Ri is a (m + 1)tuple (Oi, P (1) i , P (2) i , . . . , P (m) i ), Oi is any one in a set O = {O1, O2, . . . , Oo} of o classes, and P (1) i , P (2) i , . . . , P (m) i are m potential predictors for Oi. Our purpose is to find a statistical machine learning algorithm (SMLA) A such that Vi = A(P (1) i , P (2) i , . . . , P (m) i ), where Vi is a predicted class by A that was developed using n ≤ m correct number of predictors forO ∈ O, and A is the best algorithm such that the metric v−1 �v i=1 |Oi−Vi| is minimum across v < r records in the validation set V ⊂ R. Our problem is to find the subset {P (1) i , P (2) i , . . . , P (n) i } and to train A using t < r records from the training set T ⊂ R, such that T ∩ V = ∅, so that A can predict whether a WIFS trying to enter an undergraduate program at UPLB will incur at least a “delinquency” once the student is accepted into the program. The A can be a useful decision-support tool for UPLB deans and college secretaries in deciding whether a WIFS will be accepted into the program or not. The potential predictors P (1) i , P (2) i , . . . , P (m) i are the ith WIFS own UPCAT record such as gender, age, high school grade, province, UPCAT score, etc. In this problem, m = 21. The setO is composed of o = 5 classes, the first four of which are considered by the administration as “delinquencies.” These classes are: (1) The student will transfer to another UP campus after being accepted into a program (O1); (2) The student will incur poor scholastic performance in the program (O2); (3) The student will shift to a different program (O3); (4) The student will commit absence without leave or file for leave of absence or honorable dismissal (O4); and (5) The student will continue with the program (O5). The desirable predicted class using any A should be O5 where the decision for student acceptance into the program becomes a trivial one. Based on UPLBs freshmen intake record of AYs 2011-2012, 2012-2013, and 20132014 furnished by the Office of the University Registrar (OUR), r = 2, 302. The dataset, however, is heavily inbalanced in favor of O1 comprising about 59% ofR, which means that every 3 of 5 WIFS chose to transfer to another UP campus after having been accepted into the program, seemingly using the program as a stepping stone to the campus that the WIFS did not qualify to. The rest are 5%, 2%, 1%, and 33% for O2, ∗Submitted as a scientific oral paper contribution to the 18th National Student-Faculty Conference on the Statistical Sciences, SEARCA, Los Baños, Laguna, 16 October 2017.

A nomenclature for the standard linguistic description of the kinematics of networks

The rate of change ∂M/∂t of some metricM measured as one of the kinematic properties of a network described by a graph G transitioning from G(Vt, Et) to G(Vt+∂t, Et+∂t) over time range ∂t has been described in the literature with linguistic descriptions that often provide ambiguity. For example, one rate of change (∂M/∂t)1 has been described as “dynamic” and another (∂M/∂t)2 as “highly dynamic” but (∂M/∂t)1 > (∂M/∂t)2. We propose in this paper a nomenclature for the standard linguistic description of the kinematics of networks in the hope that description in the literature will be standardized and understood with the corresponding quantitative meaning. We termed a network as “static” when ∂M/∂t = 0, as “non-volatile” when 0 1. In the development of the linguistic nomenclature, we borrowed heavily from the standard used in signal theory to provide linguistic descriptions to various ranges for ∂M/∂t > 1. We described the kinematics of example real-world networks where the proposed nomenclature was used: (1) The collaboration network of Filipino Computer Scientists; (2) The network created from friendship relations among Batangas and Laguna Facebook users; and (3) The network created from the followed-follower relations among the top ten globally influential Twitter users. keywords: Network kinematics, nomenclature, standard linguistic description ∗Contributed abstract to the Contributed Abstract to the 2015 Mathematical Society of the Philippines Annual Convention, Plaza del Norte Hotel and Convention Center, Laoag City, 18-19 May 2015. 1 PeerJ PrePrints | https://dx.doi.org/10.7287/peerj.preprints.1374v1 | CC-BY 4.0 Open Access | rec: 17 Sep 2015, publ: 17 Sep 2015 P re P rin ts

Quantum Trajectories with Dynamic Loop Scheduling and Reinforcement Learning

The study of many problems in quantum mechanics is based on finding the solution to the time-dependent Schrodinger equation which describes the dynamics of quantum-mechanical systems composed of a particle of mass m moving in a potential V. Based on the hydrodynamic interpretation of quantum mechanics by Bohm (1952), an unstructured grid approach, the quantum trajectory method (QTM) was developed by Lopreore and Wyatt (1999). Derivatives needed for updating the equations of motion are obtained using curve-fitting by a moving weighted least squares algorithm, and analytically differentiating the least squares curves. The calculations involve computationally-intensive parallel loops with nonuniform iterate execution times

Performance evaluation of a dynamic load-balancing library for cluster computing

The performance of scientific applications in heterogeneous environments has been improved with the research advances in dynamic scheduling at application and runtime system levels. This paper presents the performance evaluation of a library as a result of an integrated approach to dynamic load balancing. This approach combines the advantages of optimising data migration via novel dynamic loop-scheduling strategies with the advances in resource management and task migration capabilities offered by a recently developed parallel runtime system. The performance of the library has been investigated by its use in three scientific applications: the N-body simulations, the profiling of automatic quadrature routines and the solution to the 3D heat equation. The investigations focus on the performance degradation owing to the overhead induced by the runtime system software layer. The experimental results obtained indicate only a slight increase in the cost of load balancing owing to this overhead. The results validate the suitability of the runtime system as an implementation platform for dynamic load-balancing schemes and underscore the significance of using the integrated approach, as well as the benefits of using the library especially in cluster applications characterised by irregular and unpredictable behaviour.