Sriram Vajapeyam

Freelance Researcher                                                  UW Alumnus Webpage                                                          sriram at alumni dot cs dot wisc dot edu
Bangalore, India                                         

Ph.D.(1991)

                                                sriram_vajapeyam at protonmail.com


Current Professional Activity:

     Freelance Researcher
     Freelancing since Feb.2010.      

Chief Interests:

     R&D: Computer Systems, especially Computer Architecture and High-Performance Computing
     Product Development: Intersection of Computer Hardware and AI / Machine Learning / Deep Learning
     Education: Computer-Aided Learning and Teaching

 

 

Research Contributions:

Processors, Approximate Computing, Heterogeneous Computing, etc.
     click here ...

Processors: Trace Processors (ISCA'97), which promised a factor of 2x to 3x performance advantage over comparable superscalar processors for branch-intensive programs, in an era of just 10% - 15% typical performance improvements from micro-architectural ideas.

Approximate Computing: Non-Strict Cache Coherence for Emerging Applications(HiPC'96). Ground-breaking work in 1996 -- two decades ahead of the current spurt. This work incorporated domain-specific architectural enhancements and vertical integration from the applicaton down to the architecture.

Heterogeneous Computing's Load-Balancing: An early, first-cut Performance Model in 1994 (published HiCSS'98) -- anticipated by two decades the current need for load balancing in a system having very short latencies between heterogeneous computing elements working together on a parallel program.

Processors: Meta-Core Processors. Proposed synaptic connections between on-chip processor cores along with language-level support (~2007, @IBM, US Patent), well ahead of similar later work at MIT.

Programming Languages: Neighboring-thread control and data-sharing at the language level, efficiently supported by meta-core processors and adapted compilers. Enables hitherto inefficient paradigms such as aspect-orientation. (US Patent)

 

 

Industry Impact:

Processors, Software Fault Tolerance
     click here ...

Processors: HAL Computers, a US-based Fujitsu subsidiary, attempted to build a commercial Trace Processor (ISCA'97) over multiple years in the late 1990s. This was described in an article by a leading industry publication, Microprocessor Report ,www.mdronline.com, although the article failed to credit this author for the original proposal of Trace Processors (ISCA'97), instead crediting follow-up papers by other researchers. (The error was subsequently acknowledged in personal communication).

Diefendorff, Keith (15 November 1999). "Hal Makes Sparcs Fly". Microprocessor Report, Volume 13, Number 5.


Software Fault Tolerance: Sriram Vajapeyam's seed ideas on dynamic value-tracking and error handling in programs directly motivated Anant Agarwal of MIT into starting up InCert Corp., a Cambridge, Massachusetts based startup company in 1997, that provided transparent, on-the-fly support software for application fault-handling. The original idea was to provide an automatic assistant for addressing the Y2K problem but the company ended up addressing the mainframe reliability market instead. InCert Corp. was subsequently acquired by Geodesic Systems of Chicago, which in turn was acquired by Veritas that was then acquired by Symantec.

 

 

Technical Fascinations:

Many!
     click here ...

"Fascination" (defn.):     "I find work fascinating; I can sit and watch it for hours!" -- Dennis The Menace!


- Information Theory
- AI: Machine Learning, Deep Learning
- Privacy Preservation in Data Mining
- Quantum Computing,  ...   etc, etc.

 

 

Select Publications:

ISCA, etc.
     click here ...

Multi-core (Tiled) Processors. CDE: A Compiler-Driven, Dependence-Centric, Eager-Executing Architecture for the Billion Transistor Era , published at the Workshop on Complexity-Effective Design held at ISCA'03. Evangelised at the ISCA 2004 panel discussion on ILP architectures. [Slides]


Trace Processors, a novel micro-architecture. Published under the title " Improving Superscalar Instruction Dispatch and Issue by Exploiting Dynamic Code Sequences " in the 1997 IEEE/ACM International Symposium on Computer Architecture (ISCA-24).
Old webpage of Trace Processors project .


Dynamic Vectorization. A hardware method for transparently converting program loops to vector form during execution. Published under the title " Dynamic Vectorization: A Mechanism for Exploiting Far-Flung ILP in Ordinary Programs " in the 1999 IEEE/ACM International Symposium on Computer Architecture (ISCA-26).


Sub-tagged Caches. Motivational study published as ""Page-Level Behavior of Cache Contention ",  S. Tambat, S. Vajapeyam, in TCCA Computer Architecture Letters, July 2002. Details presented in the technical report " Subtagged Caches: Study of Variable Cache-Block Size Emulation ", S. Tambat, S. Vajapeyam,  July 2001.


"Multiprocessor Cache Coherence. Non-Strict Cache Coherence: Exploiting Data-Race Tolerance in Emerging Applications ", by S. Tambat and S. Vajapeyam, Int'l. Conf. on Parallel Processing (ICPP) August 2000.

Intellectual Property:

US Patents
     click here ...

          US Patent Number            Title

  1     US Patent 9,940,170          Dynamically managing distribution of data and computation across cores for sequential programs  

  2     US Patent 9,916,086          Content-addressable memory device  

  3     US Patent 9,864,709          Data transfer in a multi-core processor  

  4     US Patent 9,811,469          Sequential access of cache data  

  5     US Patent 9,804,896          Thread migration across cores of a multi-core processor  

  6     US Patent 9,785,568          Cache lookup bypass in multi-level cache systems  

  7     US Patent 9,483,318          Distributed procedure execution in multi-core processors  

  8     US Patent 9,405,691          Locating cached data in a multi-core processor  

  9     US Patent 8,694,962          Aspect-oriented parallel programming language extensions (@IBM)

10     US Patent 7,363,467          Dependence-chain processing using trace descriptors having dependency descriptors (@Intel)

11     Prior Art Disclosure IPCOM000199882D Trace-Core Processors. 20 Sept.2010. (Original IBM Invention Disclsoure: Nov.2008.) (@IBM)

 

 

Teachings:

  • Understanding Shannon's Entropy metric for Information
  • Tutorials:

    Processor Architecture; Virtual Machines.
         click here ...

    - ISCA'98
    - ASPLOS'98
    - ISCA'99
    - ISCA'01
    - HiPC'02.

     

     

    Guest Editorials:

    IEEE Computer, etc.
         click here ...

    "Early 21st Century Processors", by Sriram Vajapeyam and Mateo Valero, Guest Editorial, IEEE Computer, April 2001.


    "Computational Science ", by Sriram Vajapeyam and Rudra Pratap, Introduction to Special Section of  Current Science , Vol 78, No 7, 10th April 2000.

    Panel Discussions:

    ISCA, etc.
         click here ...

    ISCA'04 Panel: Panelist. Topic: Supporting ILP in tiled architectures: wasted effort, or a good idea? [Slides]


    HiPC'99 Panel: Organizer. Topic: Whither Indian Computer Science R&D?

     

     

    Visiting Positions:

    MIT, Cray Rsch., UW-Madison, Intel MRL, etc. etc.
         click here ...

    - MIT (June-Aug.'97);

    - UW-Madison (May '97);

    - Intel MRL (Feb-Mar '01, May '01);

    - Cray Research (summer 1994);

    - ADI, Bangalore (Aug.'00-Jan.'01);

    - UPC Barcelona (summers '02, '03);

    - USC-LA (summer 2000);

    - ACRI, France (summer 1993);

    Past Positions:

    Click here ...

    - Research Staff Member, IBM India Research Lab., Bangalore. (July 2006 -- Jan. 2010)
    Researched Multi-Core Processor Architectures (2009).
    Previously researched processor architecture for the Blue Gene-Q (BGQ) Supercomputer team (mid-2006 - 2008).

    - Director, Oracle Real-Time Collaboration Research Group. (Dec.2003 - Feb.2006)
    Built Oracle's first formal, tiny applied-research team. Team conducted applied research in the Real-Time Collaboration space and prototyped ideas in the context of the product code. Team was based at Oracle, Bangalore, India.


    - Indian Academia (Nov.1992 - July 2001)

    - Cray Research (Aug.1991 - Oct.1992)

     

     

    Service:

         General Co-Chair: HiPC 2000 and HiPC 2001 conferences.

     

     

    Alma Maters:

    UW-Madison, IIT-Madras, etc.
         click here ...

           University of Wisconsin, Madison



    IITM       IIT (Indian Institute of Technology), Madras



            Govinda Dasa Pre-University College, Surathkal, India



            Vidyadayinee High School, Surathkal, India



            KREC Higher Primary School, Surathkal, India






    Last updated August 2018.
         click here for updates list...

    <Updated June 2018>
    <Updated May 2018>
    <Updated August 2016>
    <Updated July 2011>

     

    Visitors Stats