Kalapriya Kannan

CrossRoads: Seamless VM mobility across data centers through software defined networking

Most enterprises today run their applications on virtual machines (VMs). VM mobility - both live and offline, can provide enormous flexibility and also bring down OPEX (Operational Expenditure) costs. However, both live and offline migration of VMs is still limited to within a local network because of the complexities associated with cross subnet live and offline migration. These complexities mainly arise from the hierarchical addressing used by various layer 3 routing protocols. For cross data center VM mobility, virtualization vendors require that the network configuration of the new data center where a VM migrates must be similar to that of the old data center. This severely restricts wide spread use of VM migration across data center networks. For offline migration, the above limitations can be overcome by reconfiguring IP addresses for the migrated VMs. However, even this effort is non-trivial and time consuming as these IP addresses are embedded in various configuration files inside these VMs. As enterprises grow and new data centers emerge in different geographic locations, there is a need to interconnect these data centers in a way that allows seamless VM mobility. In this context, we present CrossRoads - a network fabric that provides layer agnostic and seamless live and offline VM mobility across multiple data centers. We leverage software defined networking and implement an OpenFlow based prototype of CrossRoads. CrossRoads extends the idea of location independence based on pseudo addresses proposed in recent research to work with a control plane overlay of OpenFlow network controllers in various data centers. We evaluate CrossRoads on an innovative testbed that leverages nested virtualization to emulate two data centers. Our results confirm that CrossRoads has negligible performance overhead as compared to a Default layer 2 network - its average performance was no worse than 2.3% as compared to Default fabric across all experiments. In some experiments, it even outperformed the Default by up to 30%.

Compact TCAM: Flow Entry Compaction in TCAM for Power Aware SDN

Low latency lookup (typically single cycle) has made Ternary Content Addressable Memory (TCAM) indispensable in high performance network switching devices. However, high power dissipation of TCAM makes it incongruous in switches for today’s power sensitive emerging network framework, viz., Software Defined Network (SDN). In this paper we propose Compact TCAM, an approach that reduces the size of the flow entries in TCAM. We use shorter tags for identifying flows than the original number of bits used to store the flow entries for SDN switches. We leverage the dynamic programming capability of SDN to route the packets using these tags. We show that our approach can be easily implemented using the new SDN framework while optimizing the TCAM space. Our experiments with real world and synthetic traffic show average reduction of TCAM power by 80% in SDN switching devices for a given number of flows.

Tag-In-Tag: Efficient flow table management in SDN switches

Ternary Content Addressable Memory (TCAM) with O(1) look up performance has become the obvious and irreplaceable choice of high performance switching hardware. However, emerging network paradigm, especially Software Defined Networking (SDN), has changed the nature of operations and the rate of access in this memory subsystem. These conditions are expected to adversely impact TCAM power consumption, increase the silicon area and hence are likely to bring down the expected performance. In this paper we propose Tag-In-Tag an approach that exploits SDN features and replaces the flow entries with two layers of simpler and shorter tags. One level of tagging exploits the availability of unique path for individual flows from the ingress switch to egress switch that can be computed a-priori. Second level of tagging allows finer identification of the flows to enable flow specific actions. Double tagging helps in preserving the finer benefits of the SDNs while providing highest level of compaction to the flow entries in the flow tables. Through various experiments using real world and synthetic data we show that our approach can accommodate 15 times more flow entries in a fixed size TCAM whereas power consumption per-flow is reduced by 80% compared to an unoptimized SDN enabled switch.

Enhancing Asset Search and Retrieval in a Services Repository using Consumption Contexts

Software organizations wanting to implement a systematic reuse program face the challenge of organizing and cataloging their software assets so that they can be retrieved in different contexts of usage across their divisions. With the advent of services-oriented architectures (SOA), of which Web services is an example, software components are readily available as services on the web using standard protocols. However, service descriptions are usually only those that are provided by the service producers and unless a service producer has thought about all the contexts in which its service may be used, there is no guarantee that the service can be retrieved with high recall. In this paper, we investigate how different contexts of asset consumption may be used for better asset modeling and discovery. We introduce an extensible set of consumption factors where the service descriptions may be provided by the service producer or other roles, implement a prototype that can evolve with the modeled factors, and demonstrate that the solution enables improved precision and recall for services available in a large-scale software organization.

Promoting Reuse via Extraction of Domain Concepts and Service Abstractions from Design Diagrams

Systematic reuse of software artifacts has been an elusive goal for several years. Service-oriented architecture (SOA) has been touted in recent years due its promise of fostering reuse. Even so, reuse with SOA continues to be limited due to the lack of formal techniques for extracting domain knowledge from existing reusable software assets. In this paper,we present an approach that extracts the domain knowledge and service abstractions from design diagrams of existing software solutions and represents it in a form that can be reused in new projects. We have implemented our approach and preliminary results indicate that both domain knowledge and service abstraction thus extracted can promote reuse of software assets to a large extent.

FlowMaster: Early Eviction of Dead Flow on SDN Switches

High performance switches employ extremely low latency memory subsystems in an effort to reap the lowest feasible end-to-end flow level latencies. Their capacities are extremely valuable as the size of these memories is limited due to several architectural constraints such as power and silicon area. This necessity is further exacerbated with the emergence of Software Defined Networks SDN where fine-grained flow definitions lead to explosion in the number of flow entries. In this paper, we propose FlowMaster, a speculative mechanism to update the flow table by predicting when an entry becomes stale and evict the same early to accommodate new entries. We collage the observations from predictors into a Markov based learning predictor that predicts whether a flow is valuable any more. Our experiments confirm that FlowMaster enables efficient usage of flow tables thereby reducing the discard rate from flow table by orders of magnitude and in some cases, eliminating discards completely.

Modeling the impact of review dynamics on utility value of a product

Manufacturer-provided specifications often do not provide a true picture of the utility value of a product. A product’s true assessed value is the result of consumer opinion often conveyed via word of mouth. The increasing popularity of social media has led to the inevitable integration of the social platform with e-commerce sites where consumers share their opinions on products and prospective buyers seek the opinion of their peers before making a purchase. The influencing power of these social platforms has led to researchers mining these opinions and utilizing them to assess the value of the product. Consumer opinion can vary greatly and is dependent on several factors such as when the product is launched into the market, what competitors are offering and how their product is faring over time, etc. Hence, the assessed value of a product is subject to significant dynamism which if modeled accurately, can provide several business insights. Experience has taught us that accurately capturing the time at which opinions are expressed and identifying the attributes that influence these opinions play an important role in determining assessed value; our model aims to capture this information accordingly. Our experiments are based on large-scale review sets (approximately 30,000 reviews) collected from real-world portals such as Amazon, Mouthshut and IMDB. Validation using this real-world data confirms the superiority of our model. We demonstrate that the utility value when modeled as a function of time on the most valued attributes, provides business insights.

Managing Configuration Complexity during Deployment and Maintenance of SOA Solutions

Successful deployment and maintenance of enterprise SOA solutions involves configuring and managing several middleware software stacks. These middleware stacks have several hundreds to thousands of configuration attributes to be configured and managed. Ineffective Management of configuration information increases the cost, labor and time of troubleshooting configuration problems. The reason for it are two-folds: (a) configuration dependencies cut across different software stacks and there is no formal method to capture and consolidate them, and (b) incorrect configuration of attributes results in runtime exceptions. Correlating exceptions obtained during runtime to deployment/maintenance time configuration values is difficult as the context of the error is lost and is often not traceable. Current approaches to managing and troubleshooting problems associated with configuration information involve extensive human labor which comes at a premium. In this paper, we introduce Configuration Map (CM), which is an ordered set of configuration attributes for a deployable middleware component. Such a map when associated with exceptions helps reduce the cost of resolving configuration problems. We present several use cases and show from real life deployment scenarios that CM significantly reduces the cost of labor and time.

A Differential Approach for Configuration Fault Localization in Cloud Environments

Configuration fault localization is the process of identifying fault in the configuration of component(s) that is the source of failure given a set of observed failure conditions. Configuration faults are harder to detect than on/off failures as it involves analysis of the parameters that constitute the configuration. While distributed systems become more complex and interconnected, the requirements on configuration fault localization have changed. In this paper we present a new, simple but effective approach to configuration fault localization, which utilizes the difference in configuration parameters of components that share a resource. We establish a Reference Configuration State (RCS) by determining a set of non-faulty probing components for each faulty component with respect to shared resources. Performing difference in configuration of reference state with that of the faulty components localizes faulty configuration parameter. Experiments through simulations demonstrate that our approach is effective in identifying configuration faults with reduced time and increased accuracy. Our algorithm gracefully handles the complexity of the problem as the system size grows.

SemEnAl: Using Semantics for Accelerating Environmental Analytical Model Discovery

Web as a platform to integrate applications, encapsulated as web services and composed using semantic technologies, is well established. However, in many domains, there are far more applications whose description or implementation are available than those using service-enabled access. For example, in the environmental domain, analytical models (computational functions) are crucial to analyze collected sensor data, extrapolate them for uncovered but interesting settings of interest, and gain insights for action. However, using a particular analytical model may be appropriate under very specific conditions - terrain of region, weather conditions, pollutants, types of pollutant sources, data sampling rate, etc. Thus, finding a relevant analytical model for a given setting is of particular interest to environmental agencies around the world. But it is also a complex activity since there are hundreds of analytical models with numerous constraints. In this paper, we present SemEnAl, an approach to use semantic annotations derived from a domain ontology to address this discovery problem and demonstrate its effectiveness. Our experiments with search system for analytical models indicates that using SemEnAl can provide high precision, to an extent of about 100% relevant models for selected queries. The paper thus pushes the reach of semantics to new domains.