Scheduling and Monitoring of Internally Structured Services in Cloud ...
Recommend Documents
3-tier web application. â Deployed completely ... No secondary database replica may be deployed at the same host as some other secondary database replica.
The expansion of your organization's network perimeter into the cloud makes centralized ... Monitor your cloud-based inf
termed as Infrastructure-as-a-Service or IaaS in the case of cloud computing. In an attempt to ... Section 5 presents the network monitoring in. Cloud Computing.
Aug 18, 2015 - ered as job and the job is assigned to a virtual machine. (VM). In the proposed ... center that consists of homogeneous hosts (servers) is interconnected with the CMS ..... Function call for preemption. â. /. (23) Get all BE job .
Aug 18, 2015 - Department of Information Science and Technology, Anna University, Tamil Nadu 600025, India. Correspondence should be addressed to ...
Keywords: Cloud Computing, SaaS, [email protected], SLA, Monitoring, .... performance; then in [16] they propose an application monitoring architecture entitled.
Keywords: Cloud Computing, SaaS, [email protected], SLA, Monitoring, ... ing models at runtime and how they are used to monitor applications, SLA manage-.
This is managed by the Billing and Accounts component of the Service ... should be as small in size and as refined as possible. ..... will be dependent on the business models employed by Cloud infrastructure providers .... Software Architecture.
services hosted in the cloud platform. Due to the nature of the cloud, ..... This is a services platform hosted by Microsoft data cen- ters, which provides a platform ...
We address this limitation by co-optimising production and logistics planning and scheduling using holonic. Operations m
The Hong Kong University of Science and Technology. Clear Water Bay, Kowloon, Hong Kong e-mail: [email protected]. AbstractââScheduling multicast traffic ...
among the servers and results are coherently collected. ... queuing mechanisms; resources allocation, load monitoring, service-quality-based scheduling and ... develop an integrated platform for bioinformatics services, with a good definition.
May 19, 2010 - (2) Centralized scheduling with dedicated core (Cent ded): .... 8358 (Barcelona) quadcore platform, where
during the Subscription Term. b. Hosting and Data Center Facilities. The hosting and data center facilities supporting t
IP address by routing their traffic through a Network Address Translation (NAT) gateway in a public subnet? In addition,
Nov 23, 2012 - ... service needs. Architecture for cloud service broker is proposed to operate ... Agreements contains storage space, security, availability, etc.
As the user migrates from one computer terminal to another, he would like to ... By having an IP address any computer in .... experiment which is Desktop PC, Notebook and PDA to simulate the ... http://www.cooltown.hp.com HPLabs Technical.
Keywordsâ Resource Scheduling, Cloud Computing ... A simple integrate such as the web gateway over Internet accesses ... Application development.
Scheduling Algorithm, Cloud Computing, RSDC, PPDD. 1. ... a schematic language, equivalent network element modeling, results for infinite sized networks.
The cost of content storage, application hosting, computation and delivery can be reduced significantly by this technology. ... affordable price1. Scheduling ...
In section III, we concentrate on the MDP solution and on its structural properties, where the cases of a single server for all tasks and a dedicated server for each ...
applications, taxonomy and comprehensive survey of existing workflow scheduling algorithms. ... of different scientific applications in distributed systems. By and large a ..... Scheduling (DPDS) and (ii) ensemble planning algorithm and. worNflow ...
two tasks have the same local computational requirements, the ... tasks either locally or remotely. .... server settings (configurations) of a private cloud: 52.
increasing demand and benefits of cloud computing infrastructure, society can take leverage of intensive computing capability services and scalable virtualized ...
Scheduling and Monitoring of Internally Structured Services in Cloud ...
VMs and affect the placement of the service via constraints ... A monitoring data distribution architecture that provides data ... Which migrations are good? NP-complete! ... A SPARQL server is deployed as a subscriber so that the scheduler can ...
Scheduling and Monitoring of Internally Structured Services in Cloud Federations
Lars Larsson, Daniel Henriksson and Erik Elmroth {larsson, danielh, elmroth}@cs.umu.se
Where are the VMs now? • Cloud hosting: – No detailed information about current deployment – No way of specifying relationships to other VMs – No control over scheduling decisions made regarding the VMs • Cloud federations: – No control over which Cloud provider is executing each VM in federated Clouds (location unawareness)
Where should each VM be? • Possible to specify the internal relationship between VMs and affect the placement of the service via constraints – …without managing the infrastructure in detail • Constraints: intra-service inter-component relationships – Defined at the beginning of the service lifecycle – Preserved for entire service duration – Offers influence over placement decisions, but not full control
What is this talk about? • Our ongoing and early work on constraints-driven Cloud (IaaS) management – A way of defining service structure and placement constraints – A model and heuristic for scheduling in Cloud (federations) that abides by constraints – A monitoring data distribution architecture that provides data upon which the scheduler bases its decisions
Federations of Clouds
Example: What do we want? • 3-tier web application – Deployed completely in Europe – All components connected to an internal network – Front-ends accessible via external network • Conditions: – Primary and secondary database replicas may not be deployed on the same host – No secondary database replica may be deployed at the same host as some other secondary database replica • …and these conditions must be retained even as parts of the service are deployed on remote sites!
How do we get it? • Definition of service components – Component types act as templates for instances – Several instances can be instantiated of each type • Inter-component affinity and anti-affinity – Levels: {geographical, site, host} – For a given level and set of components, either requires or forbids co-placement
Constraint scope • Type- and instance-level constraints – Type scope affects instances of different types • “Primary and secondary database replicas may not be deployed on the same host” – Instance scope affects instances of all types, regardless of type • “Deployed completely in Europe” • “No secondary database replica may be deployed at the same host as some other secondary database replica”
Types Node Type
Abbr.
Service Root
Description Common ancestor for all service components.
Compute Resource
C
Compute resource, which can be connected to networks and storage units.
AA-constraint
A
Metadata for use within a scheduler to determine placement according to affinity and anti-affinity rules. Scope may either be type or instance and must be specified.
Block Storage
Sb
A mountable data storage for a Compute resource. Cf. Amazon EBS.
File Storage
Sf
Data storage which may be accessed by multiple Compute resources simultaneously. Cf. Amazon S3
Internal Network
Ni
Internal network for all underlying Compute resources and File storages.
External Network
Ne
External network connection (IP address) for the parent Compute or File storage resource.
Type Relations
Example
Scheduling (VM placement) • Schedulers create mappings of sets of VMs to host machines (or remote Clouds) that maximize some benefit function (e.g. profit, utilization, reputation) • In Cloud federations, remote Clouds can be regarded as logical hosts with different characteristics (e.g. network connectivity/topology and bandwidth) • The general problem is NP-complete
Model for scheduling • • • • • •
V: set of all VMs B: benefit gained from deploying the VMs in V H: set of host machines (including remote sites) M: set of mappings mv,h C: cost function of a mapping M S: estimated costs due to risk of SLA violations in migrating from one mapping to another
• Goal is to make a new mapping M that maximizes benefit after subtracting cost and potential penalties due to SLA violations
Constraints-driven heuristics • Key is to modify mappings, i.e. perform migrations – Which migrations are good? NP-complete! • AA-Constraints help us define a heuristic: – If we migrate a VM that has affinity to others, we must move them as well – Anti-affinities prevent certain migrations (or cause a series of migrations of other VMs) • SLA violation risk can be assessed using: – Long-term monitoring data to predict spikes – Short-term monitoring of VM activity – Estimation of total data transfer of VM migrations • We use this to suggest only such migrations (changes of mappings) that have a low risk of violating SLAs for sets of VMs that are related due to AA-constraints
Monitoring • Scheduling requires pertinent up-to-date information • Contemporary monitoring systems are incompatible, which is troublesome for Cloud federations – Semantic metadata can help remove this technical barrier! • We introduce the Medici monitoring data distribution architecture: – Plugins translate specific data formats from underlying monitoring systems – Designed for scalability – Asynchronous Publish/Subscribe – Designed to handle both private and public data
Medici architecture
Medici architecture • Distribution hub uses Google’s PubSubHubbub technology to notify subscribers of when new data is available – Data is presented as an Atom feed with semantic metadata extensions to a format whose content is based on that of libvirt • A SPARQL server is deployed as a subscriber so that the scheduler can make queries to it – The server can subscribe to data from remote sites as well and thus give the scheduler information from remote sites in a familiar format
Summary • Service structure and constraints give a reasonable amount of control to the service provider regarding scheduling decisions • A scheduling model where decisions are influenced by AA-constraints and monitoring data • Medici adds semantic metadata to bridge technical gaps caused by incompatibility in Cloud federations
Future work directions • Investigate a larger set of constraints for service structure than AA-constraints • Quantify benefit of using this scheduling model compared to others • Formalize and evaluate the heuristic outlined here • Validate scalability property of the monitoring architecture – Determine reasonable sizes of collated data sets
Thank you for your attention! • Questions?
Service Representation • Parts of services (compute nodes and file storages), may have different Affinities affecting the placement – Affinity may be geographical or relate to other components in the service – Anti-affinity is an unwanted relation and follows the same patterns as Affinity – We call the union of these AA-constraints • AA-constraints have two different scopes – Type scope affects instances of different types – Instance scope affects instances of all types