My work falls into two main bins: improving big data analytics and improving network resource allocation in datacenters.
Recent Papers (All)
MSR, Bing, UIUC, Steklov Math Inst.
We show how to improve the tail latency of datacenter services which are built as workflows over many components by appropriately allocating additional resources across the various stages in the workflow. Also, even a small amount of incompleteness (i.e., returning partial results) can improve latency by a lot if used well.
UC Berkeley, MSR
Some network traffic in DCs does not care about where it is sent modulo some constraints. We consider using such traffic to better balance network load.
MSR, Bing, GNS, UIUC
We show that given responsive networks and responsive applications adapting who gets to send how much, when, and along which network paths can improve network utilization without losing out on business priorities.
MSR, UC Berkeley
Can we make a switch with a few (4-8) traffic shapers (WFQs, Priority Queues) mimic the shaping behavior of a switch that has 100s-1000s of shapers?
Is there a market of data breach insurance, i.e., one that risk assesses users and businesses, incents better behavior from both and, pays out damages to recover from accidental data breaches?
MSR, Bing, UCBerkeley
This ships in Bing's Cosmos clusters since December 2011.
How would execution plans for jobs in big data clusters change if given additional information about properties of the user code, data and how the code and data interact? Can we extract such properties at scale?
PACMan: Coordinated Memory Caching for Parallel Jobs
Ganesh Anantharanayanan, Ali Ghodsi, Andrew Wang, Dhruba Borthakur, Srikanth Kandula, Scott Shenker, Ion Stoica
How to build an input cache spanning a cluster of machines to speed-up parallel executions? Hint: Not LRU.
Industrial Track Paper
This describe how we gather statistics about code and data and use them to improve execution plans in SCOPE.
MSR, Brown, Bing
How to schedule data-parallel jobs given per-job deadlines?
MSR, Univ. of Washington
How to use dynamically available links to relieve congestion off the network core? How to build such links using 60GHz devices?
UCBerkeley, MSR, Bing
Popularlity skew in input datasets slows down jobs, and a fix.
MSR, Cornell, Azure, Bing
Adversarial network control, i.e., despite the traffic profile used by selfish or misbehaving tenants, how to ensure desired network allocation across entities?
Minlan Yu, Albert Greenberg, Dave Maltz, Jennifer Rexford, Lihua Yuan, Srikanth Kandula, Changhoon Kim
The many ways in which a vanilla network config is unsuitable for datacenter applications.
Alan Shieh, Srikanth Kandula, Emin Gun Sirer.
Can we programmatically process a non-trivial fraction of packets in the middle of the network? Then what?
Ganesh Ananthanarayanan, Srikanth Kandula, Albert Greenberg, Ion Stoica, Yi Lu, Bikas Saha, Ed Harris.
This ships in Cosmos servers since May 2010. Tech. Report: MSR-TR-2010-69. Slides.
Ang Li, Xiaowei Yang, Srikanth Kandula, Ming Zhang
Hossein Falaki, Dimitrios Lymberopoulos, Ratul Mahajan, Srikanth Kandula, Deborah Estrin
Zhicheng Liu, Bongshin Lee, Srikanth Kandula, Ratul Mahajan
Alan Shieh, Srikanth Kandula, Albert Greenberg, Changhoon Kim
Hossein Falaki, Ratul Mahajan, Srikanth Kandula, Dimitrios Lymberopoulos, Ramesh Govindan, Deborah Estrin
Alan Shieh, Srikanth Kandula, Albert Greenberg
Srikanth Kandula, Jitu Padhye, Victor Bahl
Srikanth Kandula, Ratul Mahajan.
Software built as part of this work is available here.
Srikanth Kandula, Sudipta Sengupta, Albert Greenberg, Parveen Patel, Ronnie Chaiken. Talk@IMC.
Srikanth Kandula, Ratul Mahajan, Patrick Verkaik, Sharad Agarwal, Jitu Padhye, Victor Bahl.
Albert Greenberg, Navendu Jain, James Hamilton, Srikanth Kandula, Changhoon Kim, Parantap Lahiri, Dave Maltz, Parveen Patel, Sudipta Sengupta
I've worked with some amazing interns at MSR.
Srikanth Kandula is a Researcher at Microsoft Research. His research interests span many aspects of networked systems including datacenters, network management, diagnosis, applied statistical inference and security. He has published over 15 papers in top-tier venues such as SIGCOMM, NSDI, and MobiSys. He is a winner of the NSDI best student paper award (2005). He obtained his Ph. D. from the Massachusetts Institute of Technology (2008).