Microsoft internships, etc.

[Thomas Reps <reps@xxxxxxxxxxx>]

>> From trishulc@xxxxxxxxxxxxx Fri Jan  7 16:20:30 2000
>> From: Trishul Chilimbi <trishulc@xxxxxxxxxxxxx>
>> To: "'reps@xxxxxxxxxxx'" <reps@xxxxxxxxxxx>
>>
>> Hi Tom.  Happy new year.
>>
>>    The newly formed Measurement and Performance group at Microsoft Research
>> is
>> trying to identify outstanding students interested in the general area of
>> performance
>> measurement, analyses and optimization. We have openings for summer interns
>> and
>> permanent researcher positions. Some information about the group is included
>> below (our web page is still under construction).
>>
>> I would sincerely appreciate your assistance in identifying graduate
>> students that are either available as student interns or will be completing
>> their Ph.D. in the next year or so.  Please also let your students know
>> about these opportunities.
>>
>> I ask that interested students contact me directly at
>> trishulc@xxxxxxxxxxxxxx
>> Undergraduates and graduate students in other research areas should contact 
>> the recruiter for Microsoft Research at msrjobs@xxxxxxxxxxxxxx
>>
>> Thank you very much for your help.
>>
>> -Trishul
>>
>> Measurement and Performance (MAP) is a new group whose charter is to do
>> research that will better enable us to measure, understand, and improve the
>> performance and behavior of computer systems.  
>>
>> Measurement:
>> Existing systems provide too little infrastructure to understand how and why
>> applications behave the way they do. Furthermore, what tools do exist often
>> fail to communicate the right information at the right level of abstraction.
>> We want to shift the process of building high performing systems from one
>> based on experience and intuition to a more effective one based on
>> experimentation and observation.  
>>
>> Performance:
>> Measurement can be used to improve performance in two ways.  First, it
>> informs us about performance bottlenecks in existing systems, providing
>> opportunities for optimization and tuning.
>> Feedback-directed optimization is a major focus of our efforts in this area.
>> Specifically, by leveraging infrastructure that accurately tells us how,
>> when, and why bottlenecks occur, we believe that we will be able to
>> effectively predict, detect, and avoid them by taking appropriate
>> countermeasures.  Second, measurement allows us to model and predict the
>> performance of systems before they are implemented, thus informing the
>> design process to avoid bottlenecks altogether.
>>
>> Approach:
>> We take a broad view of measurement and plan to integrate data gathered from
>> many levels of a system (e.g., from the micro-architecture to the network)
>> using different approaches to instrumentation (e.g., using statistical
>> sampling and traces).  Our view of performance includes whatever overheads
>> and resource constraints make systems less effective in use.   As a result,
>> metrics of interest include computation time, micro architecture usage,
>> memory usage, throughput, I/O overhead, request/response latency, user
>> interaction times, and power consumption.
>>
>> Opportunity:
>> We plan to apply our approach across a range of systems including hand-held
>> devices, desktop machines, and server clusters.  Small, low power devices
>> present interesting challenges because they have highly constrained
>> resources and have not been carefully studied.  Multi-threaded server-side
>> applications demand high performance and scalability, and we believe that
>> data locality optimizations in these applications are likely to be
>> effective.  Distributed systems and component-based architectures introduce
>> many new problems with respect to both understanding and improving
>> performance.  
>>
>> Current members:
>> Ben Zorn
>> Trishul Chilimbi
>> Scott McFarling
>> John DeTreville
>>
>> Current projects:
>> Low overhead instrumentation
>> Data locality optimizations
>> Binary matching algorithms
>> Code layout optimizations
>> Low power optimizations
>>
>>
>>