Coded Computation Over Heterogeneous Clusters

Various types of scalable and high performance computing systems are developed to realize distributed applications. In order to realize not only scalable but also high performance information systems, serverclusterc systems are widely used. In server cluster systems, the larger electric energy is consumed since application processes are performed on the larger number of servers. The improved delay time-based (IDTB)algorithm is discussed to select a server for each request process in a homogeneous server cluster so that the total energy consumption of the homogeneous server cluster to perform application processes can be reduced. In a homogeneous cluster, every server follows the same computation model and power consumption model. In heterogeneous cluster, servers follow different types of computation models and power consumption models. In this paper, we propose an extended IDTB (EIDTB) algorithm to reduce the total energy consumption and response time of each process in heterogeneous server cluster. We evaluate the EIDTB algorithm compared with the IDTB algorithm in a heterogeneous server cluster. In the evaluation, we show the average total energy consumption of a heterogeneous server cluster in the EIDTB algorithm can be reduced maximum 57% of the IDTB algorithm.

This paper presents a natural and efficient implementation for the classical broadcast message passing routine which optimizes performance of Ethernet based clusters. A simple algorithm for parallel matrix multiplication is specifically designed to take advantage of both, parallel computing facilities (CPUs) provided by clusters, and optimized performance of broadcast messages on Ethernet based clusters. Also, this simple parallel algorithm proposed for matrix multiplication takes into account the possibly heterogenous computing hardware and maintains a balanced workload of computers according to their relative computing power. Performance tests are presented on a heterogenous cluster as well as on a homogeneous cluster, where it is compared with the parallel matrix multiplication provided by the ScaLAPACK library. Another simple parallel algorithm is proposed for LU matrix factorization (a general method to solve dense systems of equations) following the same guidelines used for the parallel matrix multiplication algorithm. Some performance tests are presented over a homogenous cluster.

In fault-tolerant information systems, application processes have tobe reliably performed in presence of server faults. Server clustersystems are widely used to realize not only high performance but alsofault-tolerant service. In order to reliably perform applicationprocesses, multiple replicas of each application process are performedon multiple servers in a server cluster system. On the other hand, thelarger amount of electric energy is consumed in a server clustersystem since multiple replicas of each application process areperformed on multiple servers. The redundant delay time based (RDTB) algorithm is discussed to select multiple servers to redundantlyperform each application process in a homogeneous server cluster sothat the total energy consumption of the server cluster can bereduced. In a homogeneous cluster, every server follows the samecomputation model and power consumption model. On the other hand, servers follow different types of computation models and powerconsumption models in a heterogeneous server cluster. Hence, a loadbalancer cannot precisely estimate the processing load and energyconsumption of each server in a heterogeneous server cluster by theRDTB algorithm. In this paper, we propose an improved RDTB (IRDTB) algorithm to select multiple servers for redundantly performing eachapplication process so that the total energy consumption of a servercluster and response time of each process in a heterogeneous servercluster can be reduced. We evaluate the IRDTB algorithm compared withthe RDTB algorithm in a heterogeneous server cluster. In theevaluation, we show the average total energy consumption of aheterogeneous server cluster in the IRDTB algorithm can be reducedmaximum 44% of the RDTB algorithm.

Heterogeneous clusters that include accelerators have become more common in the realm of high performance computing because of the high GFlop/s rates such clusters are capable of achieving. However, heterogeneous clusters are typically considered hard to program as they usually require programmers to interleave architecture-specific code within application code. We have extended the Charm++ programming model and runtime system to support heterogeneous clusters (with host cores that differ in their architecture) that include accelerators. We are currently focusing on clusters that include commodity processors, Cell processors, and Larrabee devices. When our extensions are used to develop code, the resulting code is portable between various homogeneous and heterogeneous clusters that may or may not include accelerators. Using a simple example molecular dynamics (MD) code, we demonstrate our programming model extensions and runtime system modifications on a heterogeneous cluster comprised of Xeon and Cell processors. Even though there is no architecture-specific code in the example MD program, it is able to successfully make use of three core types, each with a different ISA (Xeon, PPE, SPE), three SIMD instruction extensions (SSE, AltiVec/VMX and the SPE's SIMD instructions), and two memory models (cache hierarchies and scratchpad memories) in a single execution. Our programming model extensions abstract away hardware complexities while our runtime system modifications automatically adjust application data to account for architectural differences between the various cores.

This paper proposes a new planning method for interconnected smart distribution networks based on total supply capability (TSC), which provides better utilization of existing assets. The specific five steps of the proposed TSC-based planning are: 1) analysis of the power supply problem; 2) calculation of the TSC value; 3) verification and adjustment of TSC and total load; 4) verification between distributed TSC and each feeder load; and 5) N-1 simulation. The load allocation problem is solved with a model that is capable of searching optimized load allocation considering N-1 security constraints. Two planning cases are presented to illustrate this method and to compare it with traditional substation-network two-step planning. It is shown that the TSC-based planning can exploit existing networks efficiently and save money. The proposed planning approach is suitable for urban smart distribution networks featured by distribution automation and large-scale interconnection.

Size and composition of clusters produced by adiabatic expansion of binary gas mixtures (Ar-Kr, Kr-Xe, and N2-Ar) with various component concentrations are studied by using electron-diffraction technique. The resulting homogeneous and heterogeneous clusters are shown to have compositions substantially different from those of the primary gas mixtures and dependent on cluster size. We have found that the key parameters needed for an analysis of cluster composition are the critical cluster radius and the heavier component concentration in the gas mixture that can be used to establish the regions of existence of homogeneous and heterogeneous clusters. These critical values determine the coefficient of the enrichment of clusters with the heavier component with respect to its concentration in the primary gas mixture. Theoretical relations are obtained which provide a good quantitative description of the experimental results and can be expected to be also valid for finding the composition of binary clusters of other van der Waals systems.

Clusters, composed of symmetric multiprocessor (SMP) machines and heterogeneous machines, have become increasingly popular for high‐performance computing. Message‐passing libraries, such as message‐passing interface (MPI) and parallel virtual machine (PVM), are de facto parallel programming libraries for clusters that usually consist of homogeneous and uni‐processor machines. For SMP machines, MPI is combined with multithreading libraries like POSIX Thread and OpenMP to take advantage of the architecture. In addition to existing parallel programming libraries that are in C/C++ and FORTRAN programming languages, the Java programming language presents itself as another alternative with its object‐oriented framework, platform neutral byte code, and ever‐increasing performance. This paper presents a new parallel programming model and a library, VCluster, which implements this model. VCluster is based on migrating virtual threads instead of processes to support clusters of SMP machines more efficiently. The implementation uses thread migration, which can be used in dynamic load balancing. VCluster was developed in pure Java, utilizing the portability of Java to support clusters of heterogeneous machines. Several applications are developed to illustrate the use of this library and compare the usability and performance of VCluster with other approaches. Copyright © 2007 John Wiley & Sons, Ltd.

More and more electric systems are embedded in today aircraft. As a result, the complexity of electrical power system design is increasing and the need of generic and efficient design methods is today required. Among numerous design tasks, the allocation of electric systems on the busbars of the electrical power system is considered as an important one since it has a direct impact on the aircraft mass. But due to the high number of possible allocations and regarding the large diversity of potential sizing cases for the equipments, finding the optimal allocation of electric loads is a hard task. In this paper, the problem is formalized mathematically. Then, four stochastic optimization methods are assessed on complex load allocation problems. Based on this assessment, a genetic algorithm using niching method is considered as the most appropriate algorithm for solving this aircraft design problem.

It is challenging to execute an application in a heterogeneous cloud cluster, which consists of multiple types of virtual machines with different performance capabilities and prices. This paper aims to mitigate this challenge by proposing a scheduling mechanism to optimise the execution of Bag-of-Task jobs on a heterogeneous cloud cluster. The proposed scheduler considers two approaches to select suitable cloud resources for executing a user application while satisfying pre-defined Service Level Objectives (SLOs) both in terms of execution deadline and minimising monetary cost. Additionally, a mechanism for dynamic re-assignment of jobs during execution is presented to resolve potential violation of SLOs. Experimental studies are performed both in simulation and on a public cloud using real-world applications. The results highlight that our scheduling approaches result in cost saving of up to 31% in comparison to naive approaches that only employ a single type of virtual machine in a homogeneous cluster. Dynamic reassignment completely prevents deadline violation in the best-case and reduces deadline violations by 95% in the worst-case scenario.

Glucantime (SbV) is the first-line treatment against American Tegumentary Leishmaniasis. Resistance cases to this drug have been reported and related to host characteristics and parasite phenotypes. In this study, 12 Leishmania (Viannia) braziliensis isolates from patients that presented clinical cure (Responders—R) and relapse or therapeutic failure (Non-responders—NR) after treatment with antimony, were analyzed. These parasites were assessed by in vitro susceptibility to SbIII and SbV, serine proteases activity measured with substrate (z-FR-AMC) and specific inhibitors (TLCK, AEBSF and PMSF). In vitro susceptibility of axenic amastigotes to SbIII showed a significant difference between R and NR groups. The protease assays showed that TLCK inhibited almost 100% of activity in both axenic amastigotes and promastigotes while AEBSF inhibited around 70%, and PMSF showed lower inhibition of some isolates. Principal component and clustering analysis performed with these data yielded one homogeneous cluster with only NR isolates and three heterogeneous clusters with R and NR isolates. Additionally, differential expression of subtilisins (LbrM.13.0860 and LbrM.28.2570) and TXNPx (LbrM.15.1080) was evaluated in promastigotes and axenic amastigotes from both groups. The results showed a higher expression of LbrM.13.0860 and LbrM.15.1080 genes in axenic amastigotes, while LbrM.28.2570 gene had the lowest expression in all isolates, regardless of the parasite form. The data presented here show a phenotypic heterogeneity among the parasites, suggesting that exploration of in vitro phenotypes based on SbIII and serine proteases profiles can aid in the characterization of L. (V.) braziliensis clinical isolates.

In cluster computing, job scheduling enhances resource utilization and operational efficiency based on user requirements. Scheduling systems utilizing deep reinforcement learning have proven effective in simple scenarios, such as homogeneous clusters and single resource pools. In this paper, we extend deep reinforcement learning to more realistic scenarios: heterogeneous load-aware computing clusters, referred to as DeepHCM. DeepHCM makes scheduling decisions based on the overall state of the cluster, incorporating aggregated information about tasks and nodes. Experimental results on various simulated heterogeneous load-aware computing clusters demonstrate the effectiveness of deep reinforcement learning. Under a similar makespan, DeepHCM exhibits smaller slowdown, indicating better scheduling performance. In a cluster with three heterogeneous nodes, the slowdown was reduced from 10.461 to 8.850. DeepHCM significantly improved scheduling performance in heterogeneous load-aware clusters, particularly when resources were limited and workloads were heavy.

The cluster formation of several aromatic carboxylic acids, ferulic acid, vanillic acid, sinapinic acid, and 3,4‐dihydroxybenzoic acid was investigated by means of laser desorption into a supersonic beam followed by multiphoton ionization‐time‐of‐flight mass spectrometry. The formation of not only homogeneous clusters, but also of heterogeneous clusters with some small amino acids was studied. The different neutral clusters formed in the supersonic expansion were ionized by a multiphoton process employing either nano‐ or femtosecond laser pulses.Strong differences in the detection of cluster ions due to the laser pulse length employed for multiphoton ionization were observed. Only femtosecond activation led to mass spectra with intense signals of the cluster ions. In addition, in the case of femtosecond ionization, protonated amino acids were detected in the mass spectra. As direct ionization of the free amino acids is not possible under the chosen ionization conditions because they lack an adequate chromophore, these protonated amino acids are assumed to be formed via an intracluster proton transfer in the heterogeneous dimer and subsequent decay of the ionized cluster (dissociative proton transfer). Such well‐known processes for heterogeneous clusters consisting of a substituted aromatic molecule and small polar solvent molecules may be involved in the matrixassisted laser desorption ionization (MALDI) process.

Power management of web server clusters have become a critical design issue because of its increasing power consumption and cooling cost. Current web server clusters are normally designed to have a performance capacity to handle peak loads, where all servers are fully utilized (turned on and running at maximum frequency). But in practice, peak load conditions rarely or never happen and most of the servers are underutilized for a long time. To make a better control of the web server clusters, we propose a technique which reconfigures the cluster-based web servers by adjusting the number of active nodes and clock frequency of each node according to the load conditions to minimize the power consumption. In this paper, we focus on heterogeneous server clusters since most real-world web-server clusters consist of several homogeneous clusters and construct a heterogeneous cluster. We describe our power-performance modeling and optimization for heterogeneous cluster-based web servers for low power.

Load balancing for clusters has been investigated extensively, mainly focusing on the effective usage of global CPU and memory resources. However, previous CPU- or memory-centric load balancing schemes suffer significant performance drop under I/O-intensive workloads due to the imbalance of I/O load. To solve this problem, we propose two simple yet effective I/O-aware load-balancing schemes for two types of clusters: (1) homogeneous clusters where nodes are identical and (2) heterogeneous clusters, which are comprised of a variety of nodes with different performance characteristics in computing power, memory capacity, and disk speed. In addition to assigning I/O-intensive sequential and parallel jobs to nodes with light I/O loads, the proposed schemes judiciously take into account both CPU and memory load sharing in the system. Therefore, our schemes are able to maintain high performance for a wide spectrum of workloads. We develop analytic models to study mean slowdowns, task arrival, and transfer processes in system levels. Using a set of real I/O-intensive parallel applications and synthetic parallel jobs with various I/O characteristics, we show that our proposed schemes consistently improve the performance over existing non-I/O-aware load-balancing schemes, including CPU- and Memory-aware schemes and a PBS-like batch scheduler for parallel and sequential jobs, for a diverse set of workload conditions. Importantly, this performance improvement becomes much more pronounced when the applications are I/O-intensive. For example, the proposed approaches deliver 23.6--88.0 % performance improvements for I/O-intensive applications such as LU decomposition, Sparse Cholesky, Titan, Parallel text searching, and Data Mining. When I/O load is low or well balanced, the proposed schemes are capable of maintaining the same level of performance as the existing non-I/O-aware schemes.

PSE (Parameter Sweep Experiments) applications represent a relevant class of computational applications in science, engineering and industry. These applications involve many computational tasks that are both resource-intensive and independent. For this reason, these applications are suited for Cloud environments. In this sense, Cloud autoscaling approaches are aimed to manage the execution of different kinds of applications on Cloud environments. One of the most recent approaches proposed for autoscaling PSE applications is MIA, which is based on the multi-objective evolutionary algorithm NSGA-III. We propose to endow MIA with other multi-objective optimization algorithms, to improve its performance. In this respect, we consider two well-known multi-objective optimization algorithms named SMS-EMOA and SMPSO, which have significant mechanic differences with NSGA-III. We evaluate MIA endowed with each of these algorithms, on three real-world PSE applications, considering resources available in Amazon EC2. The experimental results show that MIA endowed with each of these algorithms significantly outperforms MIA based on NSGA-III.