The Graduate Program Committee invites you to the following PhD Defense:
Candidate: Syed Mohammad Irteza
Thesis Title: Resilient Network Load Balancing for Datacenters
Date and Time: Tuesday, December 18th, 2018 at 9:00 AM
Venue: CS Smart Lab, Ground Floor, SBASSE, LUMS
Final Defense Committee:
1. Dr. Ihsan Ayyub Qazi (CS) - supervisor
2. Dr. Zartash Afzal Uzmi (CS/EE)
3. Dr. Tariq Jadoon (EE)
4. Dr. Muhammad Hamad Alizai (CS)
5. Dr. Fahad Rafique Dogar (Tufts University)
Datacenters are the backbone for online services such as web search and social networks. To deliver high bandwidth to such services, modern datacenters are often structured as multi-rooted tree topologies, resulting in multiple paths between every pair of servers. Thus, an efficient network load balancing scheme is essential for avoiding congestion hotspots and efficiently utilizing the available bandwidth. However, designing an efficient network load balancing scheme is challenging as datacenters are prone to link failures, switch malfunctions, and traffic variations which result in congestion asymmetry across network paths.
The key contribution of this thesis is the development of per-packet load balancing schemes that perform well under such asymmetries. Our work is based on the observation that if each flow is mapped to a topology that exhibits similar congestion levels across all its paths, then flows can achieve high performance without facing packet reordering. To this end, we propose SAPS, i.e., Symmetric and Adaptive Packet Spraying that leverages recent advances in Software-Defined Networking (SDN) to create symmetric virtual topologies (SVTs), where each SVT consists of paths with similar capacity. Evaluation involving large-scale packet-level simulations and real testbed experiments show that SAPS improves upon several existing load balancing schemes under various failure scenarios.
While SAPS works well when asymmetry affects a few paths, its performance reduces to that of flow hashing based schemes (e.g., ECMP) under higher levels of asymmetry. This happens because with SAPS each link is either entirely part of a SVT or not. The second contribution of this thesis is based on the observation that modern SDN switches provide the ability to enforce rate limits on groups of flows. By leveraging this feature, we can construct SVTs by creating slices of residual capacity across paths. This forms the basis of our proposed scheme, Slices, which outperforms other network load balancing schemes (including SAPS) when most paths face asymmetry. Both schemes, SAPS and Slices, are resilient to various forms of asymmetry, are deployable on commonly available SDN switches, and perform well across different traffic workloads and traffic patterns.
Google Scholar Page: https://scholar.google.com.pk/citations?user=wHazKsgAAAAJ&hl=en
1. Syed M. Irteza, Hafiz Mohsin Bashir, Talal Anwar, Ihsan Ayyub Qazi, Fahad R. Dogar. “Load Balancing Over Symmetric Virtual Topologies”, IEEE INFOCOM 2017, Atlanta, USA, May 2017
2. Ali Munir, Ghufran Baig, Syed Mohammad Irteza, Ihsan Ayyub Qazi, Alex X. Liu and Fahad R. Dogar. “Friends, not Foes – Synthesizing Existing Transport Strategies for Data Center Networks”, ACM SIGCOMM 2014, Chicago, USA, August 2014
3. Syed Mohammad Irteza, Adnan Ahmed, Sana Farrukh, Babar Naveed Memon, and Ihsan Ayyub Qazi. “On the Coexistence of Transport Protocols in Data Centers”, IEEE ICC 2014, Sydney, Australia, June 2014
1. Syed Mohammad Irteza, Hafiz Mohsin Bashir, Talal Anwar, Ihsan Ayyub Qazi and Fahad R. Dogar. “Efficient Load Balancing Over Asymmetric Datacenter Topologies”, Computer Communications, Elsevier, 127: 1-12, September 2018
2. Ali Munir, Ghufran Baig, Syed Mohammad Irteza, Ihsan Ayyub Qazi, Alex X. Liu and Fahad R. Dogar. “PASE: Synthesizing Existing Transport Strategies for a Near-Optimal Datacenter Transport”, IEEE/ACM Transactions on Networking, 25(1): 320-334, February 2017
Syed Mohammad Irteza earned his BS and MS from LUMS in Computer Science, in 2002 and 2005 respectively. He is currently a PhD Candidate at the Department of Computer Science at LUMS, SBASSE. His research interests are in systems and networks, particularly in novel SDN inspired designs for network and transport protocols for datacenter networks. He has published in leading conferences such as ACM SIGCOMM and IEEE INFOCOM, as well as reputed journals such as IEEE/ACM Transactions on Networking. He received the best in-session presenter award at the IEEE INFOCOM 2017 at Atlanta, Georgia, USA, and the best poster award (CS department) at the LUMS SBASSE PhD Colloquium in February 2016. Irteza was an invited participant at the Microsoft Research Summer School at Cambridge, UK, in July 2014. He has taught as a lecturer at Air University and FAST-Islamabad, and as teaching fellow at LUMS. He has also spent 4 years in the software industry.
Muhammad Hamad Alizai (PhD)
Assistant Professor of Computer Science, LUMS