Portrait of Sanjeev Mehrotra

Sanjeev Mehrotra

Software Architect

About

Sanjeev Mehrotra is a Principal Software Architect and Software Engineering Manager in the Mobility and Networking Research (MNR) Group in Microsoft Research. His areas of expertise include multimedia coding, error correction coding, multimedia networking, networking for low latency applications, container orchestration platforms, and systems for AI training. He was previously the software engineering manager for the audio codecs and DSP group at Microsoft. Prior to that, he was a co-founder at VXtreme, a pioneering company in streaming media, where he worked on researching and developing video codecs.  VXtreme was acquired by Microsoft in 1997.

Sanjeev has contributed extensively to many Microsoft products as well as to open-source projects. His work has been instrumental in many shipping Microsoft products including Windows Media, Xbox, IIS Smooth Streaming, Remote Desktop, Windows Server, and Skype, as well as contributions to H.264 and MPEG-DASH.

Currently, Sanjeev is working on the following projects.

  • HybridKube: A controller, scheduler, and service framework for hybrid (e.g. a cluster spanning both edge and cloud) and multi-cluster Kubernetes deployments.
  • KubeCluster (https://github.com/sanjeevm0/kcluster): A simple, easy, and lightweight solution to deploying general purpose multi-user Kubernetes clusters providing user container isolation and quota enforcement on either bare-metal or cloud clusters.  This can be used for creating large-scale multi-user deep learning training clusters without effort.  Please ask for permission to use this repo (currently a private repo).
  • KubeGPU (https://github.com/Microsoft/KubeDevice and https://github.com/Microsoft/KubeGPU): A Kubernetes extension framework with a custom scheduler and an alternative device framework (as opposed to using device plugins) which allows for scheduling against and consuming resources in an arbitrary fashion (as opposed to a simple device count which device plugins support). For example, this allows for a scheduler to use GPU hardware topology to speed up DNN training by 2x or more.

His previous work has included the following.

  • DLWorkspace (Deep Learning Workspace): Helped create the deep learning platform which is currently in use by all of Bing.
  • Prajna: Helped write a big data analytics platform on .Net.
  • URCP (Universal Rate Control Protocol): A delay-based network rate control (congestion control) protocol used by Remote Desktop. Variation used by Skype for controlling video bitrate.
  • RemoteFX for WAN: I was the primary inventor and creator of the network protocol currently used by Remote Desktop for WAN scenarios. Currently, uses a variant of URCP with proactive intelligent forward error correction to reduce latency.
  • Error Correction Code for Skype: Created a forward error correction code for arbitrary (N,K) with guaranteed one XOR packet to significantly speed up FEC decoder.
  • Windows Server De-duplication: Wrote the core Rabin hash computation loop and helped speed it up so that it can be effectively used in the product.
  • IIS Smooth Streaming: Helped create the first prototype of what eventually became the first dynamic adaptive streaming over HTTP (DASH) solution at Microsoft.
  • Windows Media Audio: Helped create Windows Media Audio codec and in particular the low bit-rate (LBR) extensions to WMA Pro. It is still considered as one of the best sounding audio codecs even at very low bitrates, and was used by one of the first online streaming music services.
  • XMA: The audio codec used by Xbox, developed as offshoot from Windows Media Audio
  • FEC (Forward Error Correction): Developed the forward error correction library initially for video multicast. Still serves as the core FEC library within Microsoft in use by many groups
  • Windows Media Screen Codec: Developed the screen codec shipping with Windows Media.
  • H.264: Worked on video coding. Contributions which are part of the H.264 patent pool.

Sanjeev has contributed to numerous patent applications (close to 70), and published several papers. He is the winner of Microsoft Gold Star Award, and as a graduate student received the National Science Foundation Graduate Fellowship, Tau Beta Pi Fellowship, Kodak Graduate Fellowship, and a Phi Kappa Phi Fellowship (Honorable Mention).

He received his M.Sc. and Ph.D. degrees in Electrical Engineering from Stanford University in 1996 and 2000 respectively.