مشاهده مشخصات مقاله
A memory-centric approach to enable timing-predictability within embedded many-core accelerators
|
نویسنده (ها) |
-
Paolo Burgio
-
Andrea Marongiu
-
Paolo Valente
-
Marko Bertogna
|
|
مربوط به کنفرانس |
سمپوزیوم سیستمها و فنآوریهای بیدرنگ و نهفته RTEST 2015 |
|
چکیده |
There is an increasing interest among real-time systemsarchitects for multi- and many-core accelerated platforms. Themain obstacle towards the adoption of such devices withinindustrial settings is related to the difficulties in tightly estimatingthe multiple interferences that may arise among the parallelcomponents of the system. This in particular concerns concurrentaccesses to shared memory and communication resources. Existingworst-case execution time analyses are extremely pessimistic,especially when adopted for systems composed of hundreds-to-thousandsof cores. This significantly limits the potential forthe adoption of these platforms in real-time systems. In thispaper, we study how the predictable execution model (PREM), amemory-aware approach to enable timing-predictability in real-timesystems, can be successfully adopted on multi- and many-coreheterogeneous platforms. Using a state-of-the-art multi-coreplatform as a testbed, we validate that it is possible to obtain anorder-of-magnitude improvement in the WCET bounds of parallelapplications, if data movements are adequately orchestrated inaccordance with PREM. We identify which system parametersmostly affect the tremendous performance opportunities offeredby this approach, both on average and in the worst case, movingthe first step towards predictable many-core systems. |
|
قیمت |
-
برای اعضای سایت : ۱٠٠,٠٠٠ ریال
-
برای دانشجویان عضو انجمن : ۲٠,٠٠٠ ریال
-
برای اعضای عادی انجمن : ۴٠,٠٠٠ ریال
|
خرید مقاله
|
|
.gif){kind=icon}