@techreport{TR-IC-10-33,
   number = {IC-10-33},
   author = {Daniel Nicacio and Alexandro Baldassin and Guido Araujo},
   title = {{LUTS: A Lightweight User-Level Transactional Scheduler}},
   month = {December},
   year = {2010},
   institution = {Institute of Computing, University of Campinas},
   note = {In English, 22 pages.
    \par\selectlanguage{english}\textbf{Abstract}
       Transaction  scheduling  techniques  have been used as a way to
       reduce  aborts  by  avoiding conflicting transactions to run in
       parallel,  while  improving  core usage. Traditional scheduling
       techniques   are  based  on  the  operating  system  preemptive 
       scheduler,  and  thus have little or no control in choosing the
       best transactions to execute. In this paper, we propose LUTS, a
       Lightweight  User-Level  Transaction  Scheduler.  Unlike  other 
       techniques,  which  serialize  conflicting  transactions,  LUTS 
       provides  the means for selecting another transaction to run in
       parallel, thus improving system throughput. Moreover, it avoids
       most  of  the  issues  caused by pseudo parallelism, as it only
       launches   as  many  system-level  threads  as  the  number  of 
       available  processor  cores.  This is achieved by maintaining a
       set of {\em execution context records} (ECRs) which are used to
       encapsulate  the  state  of the threads. ECRs are inserted into
       the  scheduler  queue and are selected for execution, according
       to  its  conflicting  history with the running transactions. We
       discuss  LUTS design and present a prototype implementation and
       integration  with a state-of-the-art STM, tinySTM. Experimental
       results,  conducted  with  the STAMP benchmark suite, show LUTS
       efficiency  when  running high contention applications, and its
       stability  as  the  number  of  threads increase. LUTS achieved
       better  performance  results  than existing techniques in every
       STAMP  benchmark  program,  executing programs up to 1.69 times
       faster.
  }
}