Page 19 - profile2014.indd
P. 19
Computer Systems Lab
be used by cloud gaming developers, cloud service providers, widely adopted because of its utility in exploiting distributed
and system researchers for setting up a complete cloud gaming resources and processing large-scale data. Nevertheless, such
testbed. GamingAnywhere is the first open cloud gaming test- promise is diminished somewhat by the difficulty of fitting
bed to be reported in the literature. We have conducted exten- streaming data applications into MapReduce. This is because
sive experiments to quantify the performance and overhead of MapReduce is limited to the kind of applications in which every
GamingAnywhere; in addition, we have derived optimal setups input key-value pair is independent of each other. We plan to
for system parameters, which in turn allow users to install and extend the general applicability of MapReduce by enabling de-
try out GamingAnywhere on their own servers. We expect that pendence within a set of input key-value pairs. Based on this
cloud game developers, cloud service providers, system re- new model, we shall develop the corresponding methodology
searchers, and individual users will use GamingAnywhere to set and software tools capable of processing streaming applica-
up complete cloud gaming testbeds for different purposes. We tions.
firmly believe that the release of GamingAnywhere will stimu-
late further research innovation in the field of cloud gaming 4. Storage System Designs for Embedded Systems
systems, or multimedia streaming applications in general.
Embedded systems usually utilize flash memory as their stor-
3. Storage and Processing of Streaming Data in Clouds age medium. Reduced costs and advances in manufacturing
technologies have resulted in the emergence of high-density
As more and more streaming data applications are moved to multiple-level-cell and 3D flash memory chips as popular alter-
Clouds, efficient parallel frameworks and distributed file sys- natives for embedded applications; however, this has also intro-
tems are the key to meeting the scalability and performance re- duced new challenges relating to reliability, performance, and
quirements for such streaming data applications. Our research endurance. To this end, we are working on how to adopt heal-
focuses on the storage and processing of streaming data in leveling techniques to improve the endurance of 3D flash, and
Clouds. A critical requirement for the storage of streaming data are also exploring the possibility of using software solutions
is the provision of quality-of-service (QoS), which is the ability to reduce the write disturbance of real 3D flash memory. On
to guarantee a certain level of performance for an application. the other hand, we are also investigating the possibility of in-
We aim to provide QoS in distributed file systems for Clouds, troducing emerging byte-addressable, non-volatile memories,
to meet the bandwidth/latency requirement for access to each e.g., PCM, ReRAM, and STT RAM, into embedded systems. Due
QoS file, and to improve the overall utilization of storage re- to their non-volatility and byte-addressability, these emerging
sources. In the field of text processing, MapReduce has been non-volatile memories could serve as both working memory
and as a persistent store. Thus, we propose
the concept of a “one-memory system” that
adopts non-volatile memory as both its
main memory and its storage. However, ex-
isting operating systems consider storage
as block devices, and manage main mem-
ory and storage separately. In order to take
advantage of the one-memory architec-
ture, we are redesigning the memory man-
agement and storage systems of existing
operating systems. At the same time, we
are investigating new designs to optimize
the storage capacity utilization and mini-
mize redundant storage accesses, in order
to save energy and improve performance.
Using Virtual Block Remapping to Enhance Data Reliability of 3D Flash Memory.
19