Scheduling of multiple multi-threaded applications on CMPs
- Global styles
- Apa
- Bibtex
- Chicago Fullnote
- Help
Abstract
Klasik islemci tasarmndaki snrlamalardan dolay, tek bir yongada birden fazlacekirdege sahip olan Yongada C oklu _Islemciler (CMP) performans gelisimi icin tekcekirdekli mimarilere umit verici bir alternatiftir. CMP kullanm ile elde edilebilirperformans arts, coklu cokizlekli (multi-threaded) uygulamalarda paylasml onbellekyapsndaki cekismeden dolay azalabilir. Bizim esas odak noktamz, coklu cokizlekliuygulamalar icin haritalama stratejileri sunmaktr. Biz bu tezde, yeni bir tahmintabanlharitalama stratejisi sunuyor ve gelistiriyoruz. Bu yontem, farkl uygulamalarnizleklerinin paylasml onbellek uzerindeki davranslarn analiz eder, farkl uygulamalarntum izlek kombinasyonlarn tahmin eder, ve farkl uygulamalarn en az onbellek karsklgnasebep olacak en iyi izlek kombinasyonunu bulmaya calsr. Tahmin tabanl cercevemiziniki bileseni vardr: statik bilesen ve dinamik bilesen. Tahmin surecinin egitim asamasstatik bilesende cevrimds olarak yaplr. Tahmin edilen degerler alndktan sonra, heruygulamadan kac adet izlegin ayn cekirdegi paylasabilecegi ayarlanr. _Ikinci bilesendeise, egri uydurma modeli ile iletisim kurulmas, tahmin sonuclarnn alnmas ve bu tahminsonuclarna gore en son haritalama izlemi belirlenmesi calsma zamannda yaplr.Uygulama kodu ile egri uydurma modeli arasndaki iletisim runtime modulu tarafndangerceklestirilir. Bu modul, egitim asamas icin gerekli olan bilgiyi uygulama kodundanalr, egri uydurma modeline iletir ve tahmin edilen bilgileri egri uydurma modelindenalr, uygulama koduna gonderir. Hicbir admnda programa karslmaz. Due to the limitations in the conventional processor designs, chip multiprocessors(CMPs), which have multiple cores on a single chip, are a promising alternative to singlecorearchitectures for performance improvements. The potential performance gains thatcan be achieved by the using CMPs decline when there is contention for the sharedcache structure for multiple multi-threaded applications. Our main focus is to presentmapping strategies of multiple multi-threaded applications on multicore architectures.We propose and develop a novel prediction-based mapping strategy. Our approachanalyzes thread behavior of dierent applications on the shared cache by consideringall possible thread combinations of dierent applications. It nds the best threadcombinations of dierent applications that result in minimum cache disturbance. Ourprediction-based framework has two components: a static component and a dynamiccomponent. The collection of the training data which is given to the curve tting modelas an input is done o-line at the static component. After receiving the predicted values,the threads of each application that shares the same core are arranged. Communicationwith curve tting model, receiving predicted results, and nally mapping according tothese values are done on-line at the dynamic component. The communication betweenthe application code and the curve tting model is provided by a runtime module whichcollects the training data from the application code and sends them to the curve ttingmodel and receives predicted data from the curve tting model and sends them to theapplication code. Any interference with the program is avoided at every step of theexecution.
Collections