Page 18 - 2017 Brochure
P. 18
earch Laboratories
Computer Systems Lab
The Computer Systems Lab was established in 2009. 2. Design and Implementation of a Cloud
Its primary research areas include multicore systems, Gaming System
virtualization, system software for cloud computing and
related applications, and storage designs for embedded Cloud gaming systems render game scenes on cloud
systems. servers and stream the encoded scenes to thin clients
over the Internet. The thin clients then send user inputs,
1. Auto-parallelism with Dynamic Binary from joysticks, keyboards, and mice, back to the cloud
Translation servers. With cloud gaming systems, users can: (i) avoid
upgrading their computers for the latest games, (ii) play
Parallelization is critical for multicore computing and cloud the same games using thin clients on different platforms,
computing. Hardware manufacturers have adopted many such as PCs, laptops, tablets, and smartphones, and
distinct strategies to improve parallelism in microprocessor (iii) play more games due to reduced hardware/software
design. These strategies include multi-cores, many-cores, costs. Game developers may: (i) support more platforms,
GPU, GPGPU, and SIMD (single instruction, multiple data), (ii) avoid hardware/software incompatibility issues, and (iii)
among others. However, these parallel architectures have increase net revenues. Therefore, cloud gaming systems
very different parallel execution models and several issues have attracted attention from users, game developers,
arise when migrating applications from one to another: (1) and service providers. We have developed an open cloud
application developers have to rewrite programs based gaming system, GamingAnywhere, which can be used
on the target parallel model, increasing time to market; by cloud gaming developers, cloud service providers,
(2) legacy applications become under-optimized due to and system researchers for setting up a complete cloud
under- utilization of parallelism in the target hardware, gaming testbed. GamingAnywhere is the first open cloud
significantly diminishing potential performance gain; and (3) gaming testbed to be reported, and we conduct extensive
execution migration among heterogeneous architectures testing to quantify its performance and overhead. We derive
is difficult. To overcome these problems, we developed an the optimal system parameters, which allows users to
efficient and retargetable dynamic binary translator (DBT) install and try out GamingAnywhere on their own servers.
to transparently transform application binaries among We expect that cloud game developers, cloud service
different parallel execution models. In our current work, the providers, system researchers, and individual users will
DBT dynamically transforms binaries of short-SIMD loops use GamingAnywhere to set up complete cloud gaming
to equivalent long-SIMD ones, thereby exploiting the wider testbeds for different purposes. We firmly believe that the
SIMD lanes of the hosts. We have shown that the SIMD release of GamingAnywhere will stimulate more research
transformation from ARM NEON to x86 AVX2 can improve innovations on cloud gaming systems, or multimedia
performance by 45% for a collection of applications, while streaming applications in general.
doubling the parallelism factor. We plan to extend the DBT
system by supporting more parallel architectures and 3. Efficient and Scalable MapReduce Platform
execution models. Furthermore, we will re-optimize legacy
codes by exploiting enhanced ISA features provided by new MapReduce is a programming model proposed by Google
parallel architectures. for processing and generating large data sets on clouds
with a parallel, distributed algorithm. An implementation of
Dynamic Binary Translation and Parallelism Optimization System
16 研究群 Research Laboratories
Computer Systems Lab
The Computer Systems Lab was established in 2009. 2. Design and Implementation of a Cloud
Its primary research areas include multicore systems, Gaming System
virtualization, system software for cloud computing and
related applications, and storage designs for embedded Cloud gaming systems render game scenes on cloud
systems. servers and stream the encoded scenes to thin clients
over the Internet. The thin clients then send user inputs,
1. Auto-parallelism with Dynamic Binary from joysticks, keyboards, and mice, back to the cloud
Translation servers. With cloud gaming systems, users can: (i) avoid
upgrading their computers for the latest games, (ii) play
Parallelization is critical for multicore computing and cloud the same games using thin clients on different platforms,
computing. Hardware manufacturers have adopted many such as PCs, laptops, tablets, and smartphones, and
distinct strategies to improve parallelism in microprocessor (iii) play more games due to reduced hardware/software
design. These strategies include multi-cores, many-cores, costs. Game developers may: (i) support more platforms,
GPU, GPGPU, and SIMD (single instruction, multiple data), (ii) avoid hardware/software incompatibility issues, and (iii)
among others. However, these parallel architectures have increase net revenues. Therefore, cloud gaming systems
very different parallel execution models and several issues have attracted attention from users, game developers,
arise when migrating applications from one to another: (1) and service providers. We have developed an open cloud
application developers have to rewrite programs based gaming system, GamingAnywhere, which can be used
on the target parallel model, increasing time to market; by cloud gaming developers, cloud service providers,
(2) legacy applications become under-optimized due to and system researchers for setting up a complete cloud
under- utilization of parallelism in the target hardware, gaming testbed. GamingAnywhere is the first open cloud
significantly diminishing potential performance gain; and (3) gaming testbed to be reported, and we conduct extensive
execution migration among heterogeneous architectures testing to quantify its performance and overhead. We derive
is difficult. To overcome these problems, we developed an the optimal system parameters, which allows users to
efficient and retargetable dynamic binary translator (DBT) install and try out GamingAnywhere on their own servers.
to transparently transform application binaries among We expect that cloud game developers, cloud service
different parallel execution models. In our current work, the providers, system researchers, and individual users will
DBT dynamically transforms binaries of short-SIMD loops use GamingAnywhere to set up complete cloud gaming
to equivalent long-SIMD ones, thereby exploiting the wider testbeds for different purposes. We firmly believe that the
SIMD lanes of the hosts. We have shown that the SIMD release of GamingAnywhere will stimulate more research
transformation from ARM NEON to x86 AVX2 can improve innovations on cloud gaming systems, or multimedia
performance by 45% for a collection of applications, while streaming applications in general.
doubling the parallelism factor. We plan to extend the DBT
system by supporting more parallel architectures and 3. Efficient and Scalable MapReduce Platform
execution models. Furthermore, we will re-optimize legacy
codes by exploiting enhanced ISA features provided by new MapReduce is a programming model proposed by Google
parallel architectures. for processing and generating large data sets on clouds
with a parallel, distributed algorithm. An implementation of
Dynamic Binary Translation and Parallelism Optimization System
16 研究群 Research Laboratories
