This week I did a lot of reading up on cloud computing and scheduling. I read and completed 3 articles and am almost done with 2 others. I am continuing to read 3 other books: "Mining of Massive Datasets," a book about Java, and a book about C++.
Here are the three articles that I read: "Sparrow: Distributed, Low Latency Scheduling," "Load Balancing in Public Cloud by Division of Cloud Based on the Geographical Location," and "Review of Load Balancing in Cloud Computing." (Please see the tap on the top crossbar labeled Syllabus for citations of each text).
Here's a quick overview and terminology from each complete article (Note: some definitions and phrases may be directly from the article):
1) "Sparrow: Distributed, Low Latency Scheduling." Sparrow is a scheduler. A scheduler is a software application for the allocation of jobs. Basically, a scheduler is an application that takes a job off of a queue and then decides onto which machine to place it, with the goal of achieving the shortest possible run-time. See image, below (if you have difficulty viewing this image, click on it and a larger version will appear).
The researchers demonstrated that a decentralized, randomized sampling approach provides near-optimal performance while avoiding the throughput and availability limitations of a centralized design. With implementation on a 110-machine cluster, the researchers demonstrated that Sparrow performs within 12% of an ideal scheduler.
Terminology:
Worker Machines execute tasks.
Schedulers assign tasks to worker machines.
A job consists of m tasks that are each allocated to a worker machine. Jobs can be handled by any scheduler.
If a worker machine is assigned more tasks than it can run concurrently, then it will queue new tasks until it has enough available resources to handle the next one.
Wait time is the time from which a task is submitted to the scheduler to when the task begins executing.
Service time is the amount of time that the task spend executing on a worker machine.
Job Response Time is the amount of time from when the job is submitted to the scheduler to when the last task within the job is finished executing.
Delay is the total delay within a job due to both scheduling and queuing.
The next two articles reviewed Load Balancing within Cloud Computing
2) "Load Balancing in Public Cloud by Division of Cloud Based on the Geographical Location." Load Balancing is a method of controlling the traffic in a cloud environment. Effective load balancing results in the efficient allocation of resources to all competing jobs. The researchers describe load balancing in a public cloud by partitioning the cloud into several sub-clouds. This division is done by geographical location. A Central Controlling System (CCS) monitors each of the sub-clouds. Every sub-cloud has a balancing system which monitors and allocates resources to competing jobs and relays information back to the CCS. From this information, the CCS can choose the optimal sub-cloud.
3) "Review of Load Balancing in Cloud Computing." This paper gave an overview of various aspects of cloud computing, its evolution, and its issues, particularly issues related to load balancing. The paper focused on one issue of load balancing: the consequences of inefficient load balancing may result in the detriment of a business's performance in the cloud environment.
This was just a sample of what I learned from these articles.
I am also beginning to learn the Scala programming language, as this is the language in which the Spark system is rooted. Hopefully, my prior experience with Java Programming Language will help with the foundations of Scala.
If you have any questions or comments please feel free to leave a comment below and I will try to get back to you as soon as possible. Same goes for any clarification of topics or concepts.
No comments:
Post a Comment