Datacenter Traffic Control: Understanding Techniques and Trade-offs

Noormohammadpour, Mohammad; Raghavendra, C.S.

doi:10.31219/osf.io/6qzxc

Cited by 14 publications

(17 citation statements)

References 127 publications

(281 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Handling rate-limiting inaccuracies: Rate-limiting is generally not very accurate, especially if done in software [34]. To deal with inaccuracies and errors, every sender has to report back to the TE server at the end of every timeslot and specify how much traffic it was able to deliver.…”

Section: Discussionmentioning

confidence: 99%

QuickCast: Fast and Efficient Inter-Datacenter Transfers using Forwarding Tree Cohorts

Noormohammadpour¹,

Raghavendra²,

Kandula³

et al. 2018

Preprint

Self Cite

View full text Add to dashboard Cite

Large inter-datacenter transfers are crucial for cloud service efficiency and are increasingly used by organizations that have dedicated wide area networks between datacenters. A recent work uses multicast forwarding trees to reduce the bandwidth needs and improve completion times of point-tomultipoint transfers. Using a single forwarding tree per transfer, however, leads to poor performance because the slowest receiver dictates the completion time for all receivers. Using multiple forwarding trees per transfer alleviates this concern-the average receiver could finish early; however, if done naively, bandwidth usage would also increase and it is apriori unclear how best to partition receivers, how to construct the multiple trees and how to determine the rate and schedule of flows on these trees. This paper presents QuickCast, a first solution to these problems. Using simulations on real-world network topologies, we see that QuickCast can speed up the average receiver's completion time by as much as 10× while only using 1.04× more bandwidth; further, the completion time for all receivers also improves by as much as 1.6× faster at high loads.

show abstract

Section: Discussionmentioning

confidence: 99%

QuickCast: Fast and Efficient Inter-Datacenter Transfers using Forwarding Tree Cohorts

Noormohammadpour¹,

Raghavendra²,

Kandula³

et al. 2018

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…and financial considerations (SLAs). These factors affect the resource assignment policies in place (Noormohammadpour & Raghavendra 2017) and accordingly users derive some notions of fairness from them. In fact, ordinary end users comprehend such notions of fairness more easily than the ones resulting from policies implemented in the stack.…”

Section: Across the Network Pathmentioning

confidence: 99%

Distributive justice in network communications

Moktan

Manner

2019

IJCNDS

View full text Add to dashboard Cite

Communication networks often have limited resources to serve all users and flows, and therefore different network resource assignment schemes have been deployed that satisfy fairness notions such as max-min fairness or proportional fairness. The justification of why one notion suits over other must be given due analysis in designing and administrating of communication networks. We explore the development of such fairness notions and resource assignment policies from philosophical and economics roots and how they translate into network communications and its layered protocol architecture. We find that resource assignment policies are not commutative, which impacts communications traffic and its bidirectional nature. We observe the effect of prioritization in assigned resource amounts under different notions of the -fairness spectrum and when the proportion of prioritized participants is varied. After an exploratory discourse, we formulate steps for designing and administrating resource allocation schemes for network communications that takes distributive justice into consideration.Resource assignment designs attempt to obtain efficiency and optimal usage of the resources. Fairness and distributive justice are quite important factors in designing network resource assignment schemes. There are various implementations of resource assignment schemes that satisfy fairness notions such as max-min fairness, proportional fairness and others. But such implementations fail to justify one particular notion over another in any particular implementation. This is due to the lack of regard to social factors and the subjective nature of fairness while developing resource assignment algorithms in communications networks. Justice and fairness are social concepts and knowing how it is conveyed in the networking world is important for anyone trying to develop resource assignment schemes in the context of communication networks.In this paper, we highlight the translation of the commonly used fairness notions in communication networks resource assignment process to the broader philosophical concepts of social justice. We identify the different notions as belonging to different schools of justice principles. We explore the prevailing instruments of prioritization of individuals in the society for assignment of resources such as price, lottery, first-come-first-served, etc. By exploring a few resource assignment areas in communication networks and some prominent works of analysing fairness there, we follow our discussion by comparing and contrasting philosophical treatment of justice and fairness in society to that in the context of communications networks.Network traffic traverse various resource assignment nodes cascaded along the network path. Such nodes of differing policies morph into an aggregated policy resulting into different assigned resource amounts, whose fairness notions are not explained by traditional notions. Such resulting notions are of significance in designing network algorithms and architecture as well as the creatio...

show abstract

“…The ultimate goal of this process is to reach and sustain a carefully defined performance objective. It is important to distinguish the main performance objectives, usually pursued by TE (optimization mechanisms); however, integration of several TE objectives into one mechanism can be unfeasible, because certain TE goals can be mutually exclusive (e.g., network performance in terms of latency/throughput and energy efficiency) [1] [4] [6]: Minimization of network congestion: This is one of the most important performance objectives in communication networks, DCNs in particular. Congestion is one of the most significant problems, which directly affects other associated performance metrics, such as packet loss, latency and jitter.…”

Section: A the Objectives Of Te Techniques For Dcnsmentioning

confidence: 99%

“…approaches in the context of DCNs As mentioned in Section I, the operational performance requirements in DCNs are fundamentally different compared to the classical transport/WAN environments, and the shortcomings of the traditional TE approaches are described as follows (more extensive general overviews can be found in [1][4] [6]): Non-optimal path computation algorithms: Available research results in [22] show that even when using MPLS-TE solution with CSPF algorithms for path computation and network resource management, increased latency was observed in numerous experiments. This was a result of combined impact of the CSPF algorithms used, autobandwidth scaling functionality (part of MPLS-TE framework), which led to continuous path changes due to automatic bandwidth adjustments on the LSPs (depending on the variations in traffic demands).…”

Section: Limitations Of the Traditional State-of-the-art Tementioning

confidence: 99%

“…In this paper, we provide an analysis of traffic engineering (TE) capabilities for Software Defined Networking (SDN) based DCN environments and highlight the advantages and shortcomings of different strategies. The purpose of this study is to provide a condensed analysis of the most important DCN-oriented TE approaches, reported in a variety of other extensive surveys [1] [2][3] [4][5] [6], but which cover a very broad range of SDN TE areas. In addition, we illustrate two methodologies for evaluating the benefits of traffic engineering capabilities in SDN controlled DCNs.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Analysis of traffic engineering capabilities for SDN-based data center networks

Pilimon

Kentis

Ruepp

et al. 2018

2018 Fifth International Conference on Software Defined Systems (SDS)

View full text Add to dashboard Cite

In the recent years more and more existing services have moved from local execution environments into the cloud. In addition, new cloud-based services are emerging, which are characterized by very stringent delay requirements. This trend puts a stress in the existing monolithic architecture of Data Center Networks (DCN), thus creating the need to evolve them. Traffic Engineering (TE) has long been the way of attacking this problem, but as with DCN, needs to evolve by encompassing new technologies and paradigms. This paper provides a comprehensive analysis of current DCN operational and TE techniques focusing on their limitations. Then, it highlights the benefits of incorporating the Software Defined Networking (SDN) paradigm to address these limitations. Furthermore, it illustrates two methodologies and addresses the scalability aspect of DCN-oriented TE, network and service testing, by presenting a hybrid physical-simulated SDN enabled testbed for TE studies.

show abstract

Datacenter Traffic Control: Understanding Techniques and Trade-offs

Cited by 14 publications

References 127 publications

QuickCast: Fast and Efficient Inter-Datacenter Transfers using Forwarding Tree Cohorts

QuickCast: Fast and Efficient Inter-Datacenter Transfers using Forwarding Tree Cohorts

Distributive justice in network communications

Analysis of traffic engineering capabilities for SDN-based data center networks

Contact Info

Product

Resources

About