In the recent years, there has been rapid adoption of virtualized and cloud-based infrastructure by telecom providers along with RAN equipment vendors, integrators, and cloud service providers, who have been moving towards open and virtualized RAN functionality. This is driven by various factors including improving hardware capital expenditure by leveraging the commodity economics of general purpose compute infrastructure for baseband processing. Virtualization also improves RAN elasticity due to better dimensioning flexibility and orchestration capabilities provided by cloud-native technologies of containerization. Further, the decoupling of baseband HW and SW enables efficient and flexible use of the general-purpose infrastructure for other workloads. This opens up the opportunity to drive the user plane functions to far edge hubs for localized URLLC use cases and enables more AI/ML penetration due to flexible dimensioning.
DU pooling provides the greatest value in C-RAN or Centralized RAN network deployments where RAN baseband equipment from many cell sites are “pooled together” into a “cloud” of general purpose processors in far edge data centers or central offices. Furthermore, through an advanced form of DU pooling called Class II pooling, a single 5G radio can now distribute the baseband processing for each user device in the cell across multiple servers.
This means that your video conference stream’s packets can be processed by a different baseband server than your rideshare driver’s navigation application data even though you are both connected to the same radio and sharing the same airwaves. It gives unprecedented elasticity and flexibility to the network to distribute (and redistribute) traffic instantly. If a baseband server becomes overloaded or fails, all impacted user sessions can be moved to a different server without any disruption to customers. And when traffic levels fall, the RAN can consolidate user sessions into fewer servers, putting the unused servers to sleep for even greater efficiency in operating the RAN.
Edge computing expands the traditional cloud architecture with additional datacenter layers that provides computation and storage closer to the end user. Adding more Edge datacenters closer to the client device, is a key enabler for the next generation mobile and IoT applications that require low latency or that produce large volumes of data. However, Edge computing requires re-envisioning the way services are deployed currently. While existing services may be replicated across wide-area cloud datacenters to improve scalability and fault tolerance, edge computing encourages partitioning service functionality by placing components or functions at the datacenter layer that best meets performance and security requirements. Further, these Edge sites are often thinly provisioned and operate in resource constrained environments as they consist of limited (possibly heterogeneous) hardware capabilities based on the types of services they are envisioned to support. Our projects advance the key technologies required to build a platform that can support the stringent SLOs (latency, availability, etc.) required by the next generation edge services - Policy/Intent driven Homing and Placement, Global-scale data management platforms, etc.
Publications:
StepNet: A Compositional Framework with Reduced Querying for Homing Complex Network Services: Azzam Alsudais, Shankaranarayanan Puzhavakath Narayanan, Bharath Balasubramanian, Zhe Huang, Eric Keller. In the IFIP/IEEE International Symposium on Integrated Network Management (IM 2021)[PDF]
MERIT: Model-driven Rehoming for VNF Chains: Muhammad Wajahat; Bharath Balasubramanian; Anshul Gandhi; Gueyoung Jung; Shankaranarayanan Puzhavakath Narayanan.In the proceedings of the 2020 IEEE International Conference on Autonomic Computing and Self-Organizing Systems (ACSOS 2020) [PDF]
A Case for Performance-Aware Deployment of Containers: Edwin F Boza, Cristina L Abad, Shankaranarayanan Puzhavakath Narayanan, Bharath Balasubramanian, Minsung Jang. In the Proceedings of the 5th International Workshop on Container Technologies and Container Clouds (WoC 2019)
SessionStore: A Session-Aware Datastore for the Edge: Seyed Hossein Mortazavi, Mohammad Salehe, Bharath Balasubramanian, Eyal de Lara and Shankaranarayanan Puzhavakath Narayanan. In the Proceedings of 4th IEEE International Conference on Fog and Edge Computing (ICFEC 2020) [PDF]
FOCUS: Scalable Search Over Highly Dynamic Geo-distributed State: Azzam Alsudais, Mohammad Hashemi, Zhe Huang, Bharath Balasubramanian, Shankaranarayanan Puzhavakath Narayanan, Eric Keller, and Kaustubh Joshi. In the 39th IEEE International Conference on Distributed Computing Systems (ICDCS 2019) [PDF]
A Model-Driven Graybox Approach to Rehoming Service Chains: Muhammad Wajahat, Bharath Balasubramanian, Anshul Gandhi, Gueyoung Jung, Shankaranarayanan Puzhavakath Narayanan. In 26th IEEE International Symposium on the Modeling, Analysis, and Simulation of Computer and Telecommunication Systems (MASCOTS 2018) [PDF]
Toward Session Consistency for the Edge: Seyed Hossein Mortazavi, Bharath Balasubramanian, Eyal de Lara, Shankaranarayanan Puzhavakath Narayanan. In USENIX Workshop on Hot Topics in Edge Computing (HotEdge 2018), Boston, MA, USA [PDF]
NodeFinder: Scalable Search over Highly Dynamic Geo-distributed State: Azzam Alsudais, Zhe Huang, Bharath Balasubramanian, Shankaranarayanan Puzhavakath Narayanan, Eric Keller, Kaustubh Joshi. In 10th USENIX Workshop on Hot Topics in Cloud Computing (HotCloud 2018) Boston, MA, USA [PDF] [Slides]
In response to the stringent latency and availability requirements of modern applications, several geo-distributed cloud datastores have emerged in recent years. Despite the presence of many tools and techniques that improve the performance of these datastores, meeting application SLAs is still challenging, given the scale of applications, and their diverse and dynamic workloads which are not entirely exposed to the datastores. In our projects, we explore the performance tradeoffs of geo-distributed datastores, along various dimensions including the consistency spectrum (Quorum-based, Causal, Entry-Consistency, etc.), user mobility and workload dynamics.
Publications:
A Drop-in Middleware for Serializable DB Clustering across Geo-distributed Sites: Enrique Saurez, Bharath Balasubramanian, Richard Schlichting, BrendanTschaen, Shankaranarayanan Puzhavakath Narayanan, Zhe Huang, Umakishore Ramachandran. In the Proceedings of 46th International Conference on Very Large Data Bases (VLDB 2020) [PDF]
MUSIC: Multi-Site Critical Sections over Geo-Distributed State:Bharath Balasubramanian, Pamela Zave, Richard Schlichting, Mohammad Salehe, Shankaranarayanan Puzhavakath Narayanan, Seyed Hossein Mortazavi, Eyal De Lara, Matti Hiltunen and Kasutubh Joshi. In the Proceedings of 40th IEEE International Conference on Distributed Computing Systems (ICDCS 2020) [YouTube] [PDF]
Karma: Cost-effective Geo-replicated Cloud Storage with Dynamic Enforcement of Causal Consistency: Tariq Mahmood, Shankaranarayanan Puzhavakath Narayanan, Sanjay Rao, TN Vijaykumar, Mithuna Thottethodi. In IEEE Transactions on Cloud Computing 2018 [ISSN: 2168-7161] [PDF]
Performance sensitive replication in geo-distributed cloud datastores: Shankaranarayanan P N, Ashiwan Sivakumar, Sanjay Rao, Mohit Tawarmalani. In International Conference on Dependable Systems and Networks (DSN), Atlanta, 2014. [PDF][PPT].
There are several challenges to achieving low latency for mobile applications -- mobile web pages are becoming complex day-by-day with rich interactive content as well as increase in Web activities using small form factor devices such as smartphones and tablets connected over high-latency cellular networks. In this project, we are developing a system to improve Mobile Web performance by achieving the right refactoring of browsing functionality between a proxy (in the edge) and a client, based on their respective strengths and techniques to make the proxy design scale to millions of users by reducing its computational overhead. Further, in a broader context beyond mobile, we seek to achieve low-latency for Web applications by exploring criticality-aware algorithms for Web object placement and caching in Content Delivery Networks (CDNs).
Publications:
Nutshell: Scalable Whittled Proxy Execution for Low-Latency Web over Cellular Networks: Ashiwan Sivakumar, Chuan Jiang, Yun Seong Nam, Shankaranarayanan Puzhavakath Narayanan, Vijay Gopalakrishnan, Sanjay G. Rao, Subhabrata Sen, Mithuna Thottethodi, T.N. Vijaykumar. In Proceedings of the 23rd Annual International Conference on Mobile Computing and Networking (MobiCom 2017), Snowbird, Utah, USA [1-min-video][PDF]
Reducing latency through page-aware management of web objects by Content Delivery Networks: Shankaranarayanan Puzhavakath Narayanan, Yun Seong Nam, Ashiwan Sivakumar, Balakrishnan Chandrasekaran, Bruce Maggs, Sanjay Rao. In proceedings of the 34th IFIP Performance and 42nd ACM SIGMETRICS 2016, Antibes Juan-les-Pins, France. [PDF]
PARCEL: Proxy Assisted bRowsing in Cellular networks for Energy and Latency reduction: Ashiwan Sivakumar, Shankaranarayanan P N, Vijay Gopalakrishnan, Seungjoon Lee, Sanjay Rao, Subhabrata Sen. In proceedings of the 10th International Conference on emerging Networking EXperiments and Technologies (ACM CoNEXT 2014), Sydney, Australia. [PDF].
User-perceived end-to-end latency of applications have a huge impact on the revenue for many businesses. For example, Amazon finds that every 100ms of latency costs 1% in sales, while Google Search found that a 400ms delay resulted in a 0.59% reduction in searches per user. Beyond e-commerce, bringing the latency under 100ms would imply that the user cannot differentiate between whether an application is running locally or is making remote requests. Service level agreements (SLAs) on these interactive applications often require bounds on the 90th (and higher) percentile latencies, which must be met while scaling to hundreds of thousands of geographically dispersed users. A unique focus area in this project is to develop techniques and build systems that help reduce end-to-end latency of large-scale, multi-tiered applications.
Related Publications:
Measuring and Characterizing the Performance of Interactive Multi-tier Cloud Applications (invited paper): Mohammad Hajjat, Shankaranarayanan Puzhavakath Narayanan, Ashiwan Sivakumar and Sanjay Rao. In proceedings of the 21st IEEE International Workshop on Local and Metropolitan Area Networks (IEEE LANMAN 2015), Tsinghua, Beijing, China. [PDF].
Dynamic Request Splitting for Interactive Cloud Applications: Mohammad Hajjat, Shankaranarayanan P N, David Maltz, Sanjay Rao, Kunwadee Sripanidkulchai. In IEEE Journal on Selected Areas in Communications, Volume 31, Issue 12, Page(s): 2722-2737, Year: 2013 [PDF].
Dealer - Application-aware Request Splitting for Interactive Cloud Applications: Mohammad Hajjat, Shankaranarayanan P N, David Maltz, Sanjay Rao, Kunwadee Sripanidkulchai. In proceedings of the 8th International Conference on emerging Networking EXperiments and Technologies (ACM CoNEXT 2012), Nice, France. [PDF][PPT].
Closer to the Cloud - A Case for Emulating Cloud Dynamics by Controlling the Environment: Ashiwan Sivakumar, Shankaranarayanan Puzhavakath Narayanan, Sanjay Rao. In NSF GENI Research and Educational Experiment Workshop, Los Angeles, CA 2012 [PDF][PPT].
Configuration navigation and change-auditing is one of the most complex yet common tasks performed by network operators on a regular basis. Our work helps network operators perform general or customized change-auditing at varying levels of granularity on the network.
Publications:
GCNav - Generic Configuration Navigation System: Shankaranarayanan P N, Seungjoon Lee and Subhabrata Sen. In proceedings of IEEE Symposium on Configuration Analytics and Automation (SAFECONFIG), Baltimore, MD. 2012. [PDF][PPT].