Scheduler Choice in Cluster Environment

Scheduler Choice in Cluster Environment Clusters have become more popular and ubiquitous and the number of processors in cluster have also increased considerably. They consist of collection of a homogeneous machines or a host of diverse computational devices which collaborate via a high speed network to execute high-performance applications. Computer industry has widely accepted that future performance increases must largely come from increasing the number of processing cores on a die. This has led to NoC processors. Efficient scheduling of high performance applications on these parallel computing systems is critical to enhance their performance and to improve system throughput. It has been proved that the problem of scheduling tasks with precedence constraints is NP-Complete [Papad, 1994]. The data flow model is gaining popularity as a programming paradigm for parallel computers. Many high-performance applications are a collection of modules which have control/data dependences among them. When the characteristics of an application is fully deterministic, including tasks execution time, size of data communicated between tasks, and task dependencies, the application can be represented by a Directed Acyclic Graph (DAG). With an increase in the number of processing units, expressing parallelism of an application has become a major challenge. Many studies have proved that designing parallel applications using both task and data parallelism is an effective approach than pure data or pure task parallel models. This mixed parallelism achieves both higher scalability and performance. Mixed parallel applications are represented as Parallel Task Graph (PTG), a graph of data parallel tasks. Understanding the importance of task scheduling on a parallel system, an attempt is made to address issues in scheduling multiple applications with the objectives of enhancing the performance of individual applications and also increasing the throughput the parallel computing system. In this thesis, we introduce two new algorithms Level Based Scheduler (LBS) and Improved Level Based Scheduler (ILBS) to schedule parallel applications represented as parallel task graph onto a cluster of multi-core processors with the objective of reducing their completion time. Both algorithms can be used both as static or hybrid schedulers. We argue that hybrid scheduler is a good scheduler choice in a cluster environment to optimize the utilization of its resources. We state that a better way to deal with multiple applications on a cluster is through adoptive space-sharing approach with a promise to benefit both the user and the cluster administrator. In a space-sharing approach, each application is given a set of processors and it is executed on these processors only. A parallel application can be run on a varied number of processors i.e. a moldable job. Hence we argue that it is good to change processor allotment for executing applications depending on the workload on cluster. To perform initial processor allotment and subsequent adaptations if required, methods to find the optimal and maximal number of processors that an application can utilize are developed. Also a novel method to share available processors among multiple competing task graphs is proposed. A framework is developed to bring together the proposed hybrid schedulers, methods to find processor requirement of each application, the scheme to share processors among multiple applicat ions and a new policy to decide processor allotment for each submitted application. Approaches to improve scheduling on a NoC processor is attempted. An approach to make any list scheduling method more time efficient to schedule a task graph on NoC is proposed and experimented. To schedule multiple applications on NoC, the number of cores and which cores to be assigned for each application must be decided. Our belief is that this job of deciding number of cores can be better performed by the joint collaboration of the user and system instead of any one doing it alone. Hence we have developed methods to find the optimal and maximal block of cores that an application can utilize which is later used to decide the actual core allotment for each application. Policies to decide how many and which cores to be assigned for each application are suggested. All the experiments in this thesis are carried out using a discrete event simulator. Benchmark task graphs are taken from different sources, from where other researchers have taken to compare their scheduler performance. The metrics makespan and efficiency of the schedule are used. The developed LBS is compared with MCPA the most widely accepted good scheduler and EMTS the recent PTG scheduler are chosen for performance comparison. The benchmark suite includes regular task graph, random task graph and few real applications task graph. For regular task graphs LBS shows in improvement in makespan by 2-9% in comparison to MCPA. But for irregular PTGs, LBS shows 4-12% performance improvement over MCPA, which is significantly higher than for regular PTGs. Since EMTS uses evolutionary methods, it generates better schedule but at the expense of more computing time. The proposed LBS performance is inferior to EMTS by around 2-7% and 2-4% for regular and random PTGs respectively. Another metric used is the efficiency which is a measure of effective utilization of resources. The efficiency of LBS is more than MCPA, but the improvement is less than that for makespan. This is attributed to the fact task allocation in MCPA leads to better utilization of processors than in L BS. Efficiency of LBS is more than MCPA by 1-3% and less than EMTS by 1-2%. Another scheduler ILBS is compared with LBS and TwoL[rauber 1998], a good method to schedule set of independent tasks. ILBS exhibits performance improvement of 2-7% over LBS and 2-10% over TwoL for regular PTGs. For random PTGs improvement is 6-12% over LBS and 4-8% over TwoL. The increased performance of ILBS for regular PTGs is attributed to the method of finding of the best possible schedule at each level. The performance of the proposed novel method of sharing processors among multiple task graphs is compared with the most recent methods suggested by Tapke et al. The new method exhibited a performance improvement of 6-9% for all categories of task graph and is maximum when the demand for the processors is relatively more than available processors. A complete framework is developed to tailor together the pieces of work carried out. The new policies suggested to decide processor allotment for each task graph show 4-7% performance improvement in average completion time of a task graph. The proposed policy also exhibits better performance for the time required to complete a set of task graphs by 4-7%. Thus the new policy is good from both user and system perspectives. The approach to make list scheduling method more time efficient to generate a schedule for a NoC processor is implemented in DLS[] method and it recorded around 20-45% improvement in execution time. The time is recorded by executing the application on the cycle accurate multi2sim simulator. The new policy proposed to decide the cores allotment for each application performs better than the best methods found in the literature by 4-20%. The issues in scheduling multiple applications on a cluster of multi-core processors and a NoC processor is addressed in this thesis. The observed performance improvement indicate the usefulness of proposed methods.

Literary Genre Essay

What is Literature? Language Department IPGK Pendidikan Teknik Based on your experience learning literature, what is literature? What are the characteristics of literature? Do you need to learn literature? †¢? Traditionally, literature is â€Å"imaginative† writing. †¢? However, the distinction between â€Å"real† and â€Å"fake† or â€Å"fact† and â€Å"fiction† isn’t always a good distinction; many classical works were non-fiction. Literature is Subjective †¢? Since the 1980’s, the â€Å"literary canon† of works – a group of works â€Å"agreed upon† to be â€Å"the best† by well-known scholars and critics, has been disputed. Why do you think the â€Å"canon† was disputed? †¢? The â€Å"Canon† excluded most works that were not by white, European males. †¢? Works of literature by women, homosexuals, and works by individuals of varied races, classes and ethnicities were marginalised. How did this happen? †¢? There are many ways of â€Å"writing† – but those in power recognised only one, formal way of â€Å"writing†, and this was given the higher value. †¢? Thus, the literary â€Å"canon† is a construct; it was fashioned by particular people for particular reasons at a particular time. †¢? There is no literary work or tradition that has value in and of itself †¦ †¢? †¦. even Shakespeare! †¢? In his era, Shakespeare was regarded as a hack! †¢? Time and circumstance has offered the value to particular text; and this â€Å"value† is a transitive term – it will change as the people in power change and are altered, and according to the context of the reading of a particular text. 10 years ago †¦. †¢? BLOGS were stupid. †¢? NOW, Iraq War Veterans’ BLOGS are considered vital historic and â€Å"literary† documents! Revisioning the Canon: †¢? All â€Å"literary† works are unconsciously rewritten by the societies that read them. Context †¢? Readers interpret literary works in the  light of their own concerns. †¢? Readers interpret literary works in the light of a given circumstances. †¢? Readers interpret literary works in the light of a given time period. The Diary of Anne Frank: †¢? Literature? Or Not? I Have A Dream: †¢? Is this speech by Martin Luther King, Jr. Literature? Or Not? Literature and â€Å"value† †¢? Each of us is constructed by experiences and backgrounds and emotions and ideas and prejudices and knowledge and lack of knowledge †¦ †¢? How we each respond to a particular text is deeply entwined with our broader prejudices and belief systems. Basic Definition of Literature †¢? Latin – litterae (plural for letter) †¢? Literally means â€Å"acquaintance with letters† Why Do We Study Literature? †¢? To obtain a window of the world and other cultures. †¢? To understand ourselves (how? ). †¢? To gain insights into a character’s inner thoughts, con? icts, aspirations etc. †¢? To actively shape culture through the active and articulate constructions of sociocultural realities. So †¦ what is literature? †¢? What constitutes a â€Å"literary† text? †¢? What qualities will help me to determine the â€Å"literariness† of a text? †¢? Read â€Å"What is Literature† by Jim Meyer for our class discussion this Thursday.

The Battle of Plataea Persian War History

The Battle of Plataea believed to have been fought in August 479 BC, during the Persian Wars (499 BC-449 BC). Armies Commanders Greeks Pausaniasapprox. 40,000 men Persians Mardoniusapprox. 70,000-120,000 men Background In 480 BC, a large Persian army led by Xerxes invaded Greece. Though briefly checked during the opening phases of the Battle of Thermopylae in August, he eventually won the engagement and swept through Boeotia and Attica capturing Athens. Falling back, Greek forces fortified the Isthmus of Corinth to prevent the Persians from entering the Peloponnesus. That September, the Greek fleet won a stunning victory over the Persians at Salamis. Concerned that the victorious Greeks would sail north and destroy the pontoon bridges he had built over the Hellespont, Xerxes withdrew to Asia with the bulk of his men. Before departing, he formed a force under the command of Mardonius to complete the conquest of Greece. Assessing the situation, Mardonius elected to abandon Attica and withdrew north to Thessaly for the winter. This allowed the Athenians to reoccupy their city. As Athens was not protected by the defenses on the isthmus, Athens demanded that an Allied army be sent north in 479 to deal with the Persian threat. This was met with reluctance by Athens allies, despite the fact that the Athenian fleet was required to prevent Persian landings on the Peloponnesus. Sensing an opportunity, Mardonius attempted to woo Athens away from the other Greek city-states. These entreaties were refused and the Persians began marching south forcing Athens to be evacuated. With the enemy in their city, Athens, along with representatives of Megara and Plataea, approached Sparta and demanded that an army be sent north or they would defect to the Persians. Aware of the situation, the Spartan leadership was convinced to send aid by Chileos of Tegea shortly before the emissaries arrived. Arriving in Sparta, the Athenians were surprised to learn that an army was already on the move. Marching to Battle Alerted to the Spartan efforts, Mardonius effectively destroyed Athens before withdrawing towards Thebes with the goal of finding suitable terrain to employ his advantage in cavalry. Nearing Plataea, he established a fortified camp on the north bank of the Asopus River. Marching in pursuit, the Spartan army, led by Pausanias, was augmented by a large hoplite force from Athens commanded by Aristides as well as forces from the other allied cities. Moving through the passes of Mount Kithairon, Pausanias formed the combined army on high ground to the east of Plataea. Opening Moves Aware that an assault on the Greek position would be costly and unlikely to succeed, Mardonius began intriguing with the Greeks in an effort to break apart their alliance. In addition, he ordered a series of cavalry attacks in an attempt to lure the Greeks off the high ground. These failed and resulted in the death of his cavalry commander Masistius. Emboldened by this success, Pausanias advanced the army to high ground closer to the Persian camp with the Spartans and Tegeans on the right, the Athenians on the left, and the other allies in the center (Map). For the next eight days, the Greeks remained unwilling to abandon their favorable terrain, while Mardonius refused to attack. Instead, he sought to force the Greeks from the heights by attacking their supply lines. Persian cavalry began ranging in the Greek rear and intercepting supply convoys coming through the Mount Kithairon passes. After two days of these attacks, the Persian horse succeeded in denying the Greeks use of the Gargaphian Spring which was their only source of water. Placed in a perilous situation, the Greeks elected to fall back to a position in front of Plataea that night. The Battle of Plataea The movement was intended to be completed in the darkness as to prevent an attack. This goal was missed and dawn found the three segments of the Greek line scattered and out of position. Realizing the danger, Pausanias instructed the Athenians to join with his Spartans, however, this failed to occur when the former kept moving toward Plataea. In the Persian camp, Mardonius was surprised to find the heights empty and soon saw the Greeks withdrawing. Believing the enemy to be in full retreat, he gathered several of his elite infantry units and began pursuing. Without orders, the bulk of the Persian army also followed (Map). The Athenians were soon attacked by troops from Thebes which had allied with the Persians. To the east, the Spartans and Tegeans were assaulted by Persian cavalry and then archers. Under fire, their phalanxes advanced against the Persian infantry. Though outnumbered, the Greek hoplites were better armed and possessed better armor than the Persians. In a long fight, the Greeks began to gain the advantage. Arriving on the scene, Mardonius was struck down by slung stone and killed. Their commander dead, the Persians began a disorganized retreat back towards their camp. Sensing that defeat was near, the Persian commander Artabazus led his men away from the field towards Thessaly. On the western side of the battlefield, the Athenians were able to drive off the Thebans. Pushing forward the various Greek contingents converged on the Persian camp north of the river. Though the Persians vigorously defended the walls, they were eventually breached by the Tegeans. Storming inside, the Greeks proceeded to slaughter the trapped Persians. Of those who had fled to the camp, only 3,000 survived the fighting. Aftermath of Plataea As with most ancient battles, casualties for Plataea are not known with certainty. Depending on the source, Greek losses may have ranged from 159 to 10,000. The Greek historian Herodotus claimed that only 43,000 Persians survived the battle. While Artabazus men retreated back to Asia, the Greek army began efforts to capture Thebes as punishment for joining with the Persians. Around the time of Plataea, the Greek fleet won a decisive victory over the Persians at the Battle of Mycale. Combined, these two victories ended the second Persian invasion of Greece and marked a turn in the conflict. With the invasion threat lifted, the Greeks began offensive operations in Asia Minor.

Definition and Examples of Ontological Metaphor

An ontological metaphor is a type of metaphor (or figurative comparison) in which something concrete is projected onto something abstract. Ontological metaphor (a figure that provides ways of viewing events, activities, emotions, ideas, etc., as entities and substances) is one of the three overlapping categories of conceptual metaphors identified by George Lakoff and Mark Johnson in Metaphors We Live By (1980). The other two categories are structural metaphor and orientational metaphor. Ontological metaphors  are  so natural and persuasive in our thought, say Lakoff and Johnson, that they are usually taken as self-evident, direct descriptions of mental phenomena. Indeed, they say, ontological metaphors are among the most basic devices we have for comprehending our experience. What is an Ontological Metaphor? In general, ontological metaphors enable us to see more sharply delineated structure where there is very little or none ... We can perceive of personification as a form of ontological metaphor. In personification, human qualities are given to nonhuman entities. Personification is very common in literature, but it also abounds in everyday discourse, as the examples below show: His theory explained to me the behavior of chickens raised in factories.Life has cheated me.Inflation is eating up our profits.Cancer finally caught up with him.The computer ​ went dead on me. Theory, life, inflation, cancer, computer are not humans, but they are given qualities of human beings, such as explaining, cheating, eating, catching up, and dying. Personification makes use of one of the best source domains we have--ourselves. In personifying nonhumans as humans, we can begin to understand them a little better.(Zoltà ¡n Kà ¶vecses, Metaphor: A Practical Introduction. Oxford University Press, 2002) Lakoff and Johnson on the Various Purposes of Ontological Metaphors   Ontological metaphors serve various purposes, and the various kinds of metaphors there are reflect the kinds of purposes served. Take the experience of rising prices, which can be metaphorically viewed as an entity via the noun inflation. This gives us a way of referring to the experience: INFLATION IS AN ENTITYInflation is lowering our standard of living.If theres much more inflation, well never survive.We need to combat inflation.Inflation is backing us into a corner.Inflation is taking its toll at the checkout counter and the gas pump.Buying land is the best way of dealing with inflation.Inflation makes me sick. In these cases, viewing inflation as an entity allows us to refer to it, quantify it, identify a particular aspect of it, see it as a cause, act with respect to it, and perhaps even believe that we understand it. Ontological metaphors like this are necessary for even attempting to deal rationally with our experiences.(George Lakoff and Mark Johnson, Metaphors We Live By. The University of Chicago Press, 1980) Mere Metaphors and Ontological Metaphors Within metaphor, a distinction can be drawn between mere and ontological metaphor; whereas the former simply associates a physical concept with a metaphysical one, the latter recognizes that all concepts resonate with possible transpositions and, as such, brings to the fore the world-making power of speaking. Furthermore, ontological metaphor structures experience as an openness to . . . movement between concepts.(Clive Cazeaux, Kant, Cognitive Metaphor and Continental Philosophy. Routledge, 2007)