Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userA Smart Background Scheduler for Storage Systems
92
0
0.0
(
0
)
Added by
Maher Kachmar
Publication date
2020
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
In todays enterprise storage systems, supported data services such as snapshot delete or drive rebuild can cause tremendous performance interference if executed inline along with heavy foreground IO, often leading to missing SLOs (Service Level Objectives). Typical storage system applications such as web or VDI (Virtual Desktop Infrastructure) follow a repetitive high/low workload pattern that can be learned and forecasted. We propose a priority-based background scheduler that learns this repetitive pattern and allows storage systems to maintain peak performance and in turn meet service level objectives (SLOs) while supporting a number of data services. When foreground IO demand intensifies, system resources are dedicated to service foreground IO requests and any background processing that can be deferred are recorded to be processed in future idle cycles as long as forecast shows that storage pool has remaining capacity. The smart background scheduler adopts a resource partitioning model that allows both foreground and background IO to execute together as long as foreground IOs are not impacted where the scheduler harness any free cycle to clear background debt. Using traces from VDI application, we show how our technique surpasses a method that statically limit the deferred background debt and improve SLO violations from 54.6% when using a fixed background debt watermark to merely a 6.2% if dynamically set by our smart background scheduler.
rate research
Read More
We present Kaleidoscope an innovative system that supports live forensics for application performance problems caused by either individual component failures or resource contention issues in large-scale distributed storage systems. The design of Kaleidoscope is driven by our study of I/O failures observed in a peta-scale storage system anonymized as PetaStore. Kaleidoscope is built on three key features: 1) using temporal and spatial differential observability for end-to-end performance monitoring of I/O requests, 2) modeling the health of storage components as a stochastic process using domain-guided functions that accounts for path redundancy and uncertainty in measurements, and, 3) observing differences in reliability and performance metrics between similar types of healthy and unhealthy components to attribute the most likely root causes. We deployed Kaleidoscope on PetaStore and our evaluation shows that Kaleidoscope can run live forensics at 5-minute intervals and pinpoint the root causes of 95.8% of real-world performance issues, with negligible monitoring overhead.
Deep Learning (DL) algorithms are the central focus of modern machine learning systems. As data volumes keep growing, it has become customary to train large neural networks with hundreds of millions of parameters to maintain enough capacity to memorize these volumes and obtain state-of-the-art accuracy. To get around the costly computations associated with large models and data, the community is increasingly investing in specialized hardware for model training. However, specialized hardware is expensive and hard to generalize to a multitude of tasks. The progress on the algorithmic front has failed to demonstrate a direct advantage over powerful hardware such as NVIDIA-V100 GPUs. This paper provides an exception. We propose SLIDE (Sub-LInear Deep learning Engine) that uniquely blends smart randomized algorithms, with multi-core parallelism and workload optimization. Using just a CPU, SLIDE drastically reduces the computations during both training and inference outperforming an optimized implementation of Tensorflow (TF) on the best available GPU. Our evaluations on industry-scale recommendation datasets, with large fully connected architectures, show that training with SLIDE on a 44 core CPU is more than 3.5 times (1 hour vs. 3.5 hours) faster than the same network trained using TF on Tesla V100 at any given accuracy level. On the same CPU hardware, SLIDE is over 10x faster than TF. We provide codes and scripts for reproducibility.
The availability of Infrastructure-as-a-Service (IaaS) computing clouds gives researchers access to a large set of new resources for running complex scientific applications. However, exploiting cloud resources for large numbers of jobs requires significant effort and expertise. In order to make it simple and transparent for researchers to deploy their applications, we have developed a virtual machine resource manager (Cloud Scheduler) for distributed compute clouds. Cloud Scheduler boots and manages the user-customized virtual machines in response to a users job submission. We describe the motivation and design of the Cloud Scheduler and present results on its use on both science and commercial clouds.
Big data systems development is full of challenges in view of the variety of application areas and domains that this technology promises to serve. Typically, fundamental design decisions involved in big data systems design include choosing appropriate storage and computing infrastructures. In this age of heterogeneous systems that integrate different technologies for optimized solution to a specific real world problem, big data system are not an exception to any such rule. As far as the storage aspect of any big data system is concerned, the primary facet in this regard is a storage infrastructure and NoSQL seems to be the right technology that fulfills its requirements. However, every big data application has variable data characteristics and thus, the corresponding data fits into a different data model. This paper presents feature and use case analysis and comparison of the four main data models namely document oriented, key value, graph and wide column. Moreover, a feature analysis of 80 NoSQL solutions has been provided, elaborating on the criteria and points that a developer must consider while making a possible choice. Typically, big data storage needs to communicate with the execution engine and other processing and visualization technologies to create a comprehensive solution. This brings forth second facet of big data storage, big data file formats, into picture. The second half of the research paper compares the advantages, shortcomings and possible use cases of available big data file formats for Hadoop, which is the foundation for most big data computing technologies. Decentralized storage and blockchain are seen as the next generation of big data storage and its challenges and future prospects have also been discussed.
The start of data taking at the Large Hadron Collider will herald a new era in data volumes and distributed processing in particle physics. Data volumes of hundreds of Terabytes will be shipped to Tier-2 centres for analysis by the LHC experiments using the Worldwide LHC Computing Grid (WLCG). In many countries Tier-2 centres are distributed between a number of institutes, e.g., the geographically spread Tier-2s of GridPP in the UK. This presents a number of challenges for experiments to utilise these centres efficaciously, as CPU and storage resources may be sub-divided and exposed in smaller units than the experiment would ideally want to work with. In addition, unhelpful mismatches between storage and CPU at the individual centres may be seen, which make efficient exploitation of a Tier-2s resources difficult. One method of addressing this is to unify the storage across a distributed Tier-2, presenting the centres aggregated storage as a single system. This greatly simplifies data management for the VO, which then can access a greater amount of data across the Tier-2. However, such an approach will lead to scenarios where analysis jobs on one sites batch system must access data hosted on another site. We investigate this situation using the Glasgow and Edinburgh clusters, which are part of the ScotGrid distributed Tier-2. In particular we look at how to mitigate the problems associated with ``distant data access and discuss the security implications of having LAN access protocols traverse the WAN between centres.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
