Co-allocation of compute and network resources using HARC

MacLaren, Jon

doi:10.22323/1.041.0016

Cited by 15 publications

(8 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The general architecture consists of Clients which generate resource coallocation requests, Acceptors which make the reservations and lastly Resource managers which talk to local schedulers for each resource. There are two important Resource Managers: (i) Compute Resource Manager which communicates with some batch scheduler (e.g, Moab, Torque or Maui) and (ii) Network Resource Manager [18] which sends commands (with the fully specified route in an Explicit Route Object) to the GMPLS controller at the path's ingress point.…”

Section: Converged Network and It Control Architecturesmentioning

confidence: 99%

NCP+: An integrated network and IT control plane for cloud computing

Buysse

Leenheer

Contreras

et al. 2014

Optical Switching and Networking

View full text Add to dashboard Cite

Cloud computing, building on the idea of "computation as a public utility" is made possible by the increased network capabilities in terms of bandwidth and reduced latency. Today, the cloud paradigm today sees adoption in many businesses, given its advantages not only from a customer point of view (e.g., universal access to the same applications across all company branches), but also from the application provider and network operator perspective (e.g., software updates no longer need to be distributed). To be able to offer cloud computing services efficiently, service providers need not just an infrastructure comprising both network and IT resources, but especially a control system that is able to orchestrate such integrated network and IT services. This paper offers a new proposal for such a system: an enhanced network control plane, the NCP+, which is based on a GMPLS control plane with a Hierarchical Path Computation Element (PCE) architecture able to jointly make network routing and IT server provisioning decisions. Indeed, in the assumed cloud paradigm, a user generally does not care what exact server the offered service is using, as long as its service requirements are met: the anycast principle applies. The paper discusses (i) the architecture of the NCP+, (ii) two IT-aware aggregation mechanisms to be used in the hierarchical PCE approach, and (iii) routing and scheduling algorithms for those aggregation mechanisms. We conclude this work with a thorough simulation analysis of the aggregation and routing/scheduling policies showing that Full Mesh aggregation where the domain topology is represented by a complete graph, although less scalable in terms of computation time, is able to provision efficiently using the proposed load balancing routing and scheduling policy. However, for a scenario with stringent IT requirements, Star could be used in parallel for scenarios where end-to-end setup times are important.

show abstract

Section: Converged Network and It Control Architecturesmentioning

confidence: 99%

NCP+: An integrated network and IT control plane for cloud computing

Buysse

Leenheer

Contreras

et al. 2014

Optical Switching and Networking

View full text Add to dashboard Cite

show abstract

“…self-organization of collaborative environment, 2. incorporation/encapsulation of external tools (even those which don't support the middleware directly), 3. continuous adaptation on changing conditions based on built-in monitoring of applications, nodes, and network links, 4. support for media streams with bitrate comparable to capacity of links (including advanced scheduling of data streams to links), 5.…”

Section: Design Principlesmentioning

confidence: 99%

“…Because the CoUniverse is designed to integrate highbandwidth applications, it is necessary to interface with services provided by advanced networks like lambda services [2] or network resource allocators [4]. The CoUniverse includes notion of a lambda link: it is a link that may be down because it is not allocated and may be allocated prior to being used, provided there is an identity participating inside the collaborative Universe, that has sufficient privileges to do so.…”

Section: Design Principlesmentioning

confidence: 99%

“…In the current implementation of CoUniverse, the Universe Peer is able to control a variety of UltraGrid flavors [9] for both uncompressed and compressed full 1080i HD video transmissions-compared to description in [9], bitrates from 250 Mbps to 1.5 Gbps are now also supported, based on several compression and bitrate reduction algorithms. Amongst other supported applications are: VideoLan Client 3 for HDV video transmissions, VIC 4 for low bandwidth videoconferencing (used as a fallback for building of the collaborative environment) and RAT 4 tool for audio transmissions.…”

Section: Prototype Implementationmentioning

confidence: 99%

See 1 more Smart Citation

CoUniverse: Framework for Building Self-organizing Collaborative Environments Using Extreme-Bandwidth Media Applications

Liška

Holub

2009

Euro-Par 2008 Workshops - Parallel Processing

View full text Add to dashboard Cite

Using internet for media-based real-time collaboration has become widespread activity. In this paper, we present a framework called CoUniverse, designed for building realtime user-empowered collaborative environments to work primarily on high-speed networks with true high-bandwidth applications such as uncompressed high-definition video. The system is designed for unreliable experimental infrastructures and therefore its operation relies heavily on selforganizing principles-this is also useful approach for extending it to larger infrastructures. When media stream bitrate is comparable to capacity of the links, the additive assumption no longer holds and the system needs to have sophisticated scheduling. Concept of scheduler is a flexible plug-in for the CoUniverse framework and in this paper, we present a formal scheduling model based on constraint programming including evaluation of its prototype implementation. CoUniverse is designed to utilize external media application, so that wide variety of existing tools can be used. The whole system has been prototyped and demonstrated, e. g., during international demonstration on GLIF 2007 workshop.

show abstract

“…A prototype Network Resource Manager (NRM) was designed for HARC that could communicate with the Calient PXCs; to set up a lightpath, a TL1 command 6 is sent to the switch at the one end of the lightpath. The HARC NRM is described in more detail in [15].…”

Section: Processing the Messagesmentioning

confidence: 99%

HARC: The Highly-Available Resource Co-allocator

MacLaren

On the Move to Meaningful Internet Systems 2007: CoopIS, DOA, ODBASE, GADA, and IS

Self Cite

View full text Add to dashboard Cite

Abstract. HARC-the Highly-Available Resource Co-allocator-is a system for reserving multiple resources in a coordinated fashion. HARC can handle different types of resource, and has been used to reserve time on supercomputers across a US-wide testbed, together with dedicated lightpaths connecting the machines. At HARC's core are a distributed set of processes called Acceptors, which provide a co-allocation service. HARC functions normally provided a majority of the Acceptors are working; this replication gives HARC its exceptional availability. The Paxos Commit protocol ensures that consistency across all Acceptors is maintained. This paper gives an overview of HARC, and explains both how it works and how it is used. We show that HARC's design makes it easy for the community to contribute new components for co-allocating different types of resource, while the stability of the overall system is maintained.

show abstract

Co-allocation of compute and network resources using HARC

Cited by 15 publications

References 5 publications

NCP+: An integrated network and IT control plane for cloud computing

NCP+: An integrated network and IT control plane for cloud computing

CoUniverse: Framework for Building Self-organizing Collaborative Environments Using Extreme-Bandwidth Media Applications

HARC: The Highly-Available Resource Co-allocator

Contact Info

Product

Resources

About