This document summarizes the PhD work of Pradeeban Kathiravelu on improving scalability and resilience in multi-tenant distributed clouds. It describes two approaches: 1) SMART uses SDN to provide differentiated quality of service and service level agreements by dynamically diverting and cloning priority network flows. 2) Mayan componentizes big data services as microservices that can be executed in a network-aware and scalable way across distributed clouds. Evaluation shows these approaches improve speedup and ensure SLAs for critical flows compared to network-agnostic distributed execution.

1. Scalability and Resilience of Multi-Tenant Distributed Clouds
in the Big Services Era
Pradeeban Kathiravelu1,2
1
INESC-ID Lisboa / Instituto Superior Técnico, Portugal. 2
Université catholique de Louvain, Belgium.
PhD Advisors: Prof. Luís Veiga, Prof. Marco Canini, Prof. Peter Van Roy.
Introduction

Data-centric big services with complex workloads.
 Geographically distributed big data to be accessed and processed.
 Resource availabilities at remote locations; for example,

Distributed clouds consist of tenants with multiple roles and locality.

Volunteer computing leverages idle resources of client computers.

Service Level Agreements (SLAs) in multi-tenant distributed clouds.
 End-to-end delivery guarantee, despite congestion, for critical flows.
Challenges

How to offer differentiated Quality of Service (QoS) in cloud networks?
 Can we discriminate the flows with redundancy in data and execution paths?
Motivation

Network-level performance guarantees based on application-level inputs.
 Scalability and Resilience for big services in the distributed clouds.
Software-Defined Networking (SDN) for Distributed Clouds

Cross-layer optimization of multi-tenant cloud networks
 Leveraging SDN and middleboxes.
Our Approach
1. SMART (SDN Middlebox Architecture for Reliable Transfers)

SMART: Network Resilience with Differentiated SLAs.
 Policies and tenant preferences: application layer → network (Figure 1).

Timely delivery of priority flows:
 Dynamically diverting them to a less congested path.
 Cloning subflows of higher priority flows.
 An adaptive approach in cloning and diverting of the flows.

An approach motivated by FlowTags Middlebox to tag the network flows.
Solution Architecture

A cross-layer architecture and communication, ensuring differentiated QoS.

A context-aware approach in load balancing the network.
 servers supporting multihoming, connected topologies, …
 Extend beyond data centers: network nodes → virtual executions.

Blurring the borders between the networks and the applications.
Contributions
Prototype Evaluation

SMART Adaptive Clone.

Improving shortest-path to
provide differentiated QoS
and SLAs in data centers.

SLA Guarantees for Critical
Network Flows (Figure 2).
* These results have been partly discussed in our recent publications: ICWS (2016), CoopIS (2015, 2016), NCA (2016), IM (2017), IC2E (2016), and SDS (2015, 2016, 2017).
Questions and Comments? pradeeban.kathiravelu@tecnico.ulisboa.pt
2. Mayan (Componentizing Data-Centric Big Services In the Internet)

An Inter-cloud framework to componentize big services.
 Execute them as a network-aware distributed service composition.
 Modelling, Scalability, and orchestration.

Use the best-fit execution path among available alternatives (Figure 3).
 Web services and microservices as the building blocks of the big services.
Solution Architecture

A scalable resilient framework for big service execution.
 Services as building blocks of the composition of big services (Figure 4).
 Message-Oriented Middleware (MOM) for inter-domain communications.
Prototype Evaluation
●
Mayan benchmarked against the
network-agnostic distributed
execution (Figure 5).
 Improved speedup for large
number of execution instances.
Contributions

Aim: Synergy of network and service level properties in decision making.

Arch: A federated controller deployment to orchestrate inter-cloud networks.

Approach: Componentizing big services as service compositions.

Contribution: Scalability and resilience for multi-tenant distributed clouds.

Future Work: Leverage cloud providers to provide differentiated connectivity.
Conclusion

Increased QoS and Speedup with network-aware scalability.
 Performance growth = f(problem size, workflow as services).

Federated deployment of SDN controller clusters.