documentclass[a4paper,12pt]{report}
usepackage{apacite}
usepackage{longtable}
usepackage{enumerate}
usepackage{amsmath}
makeatletter
renewenvironment{thebibliography}[1]{%
% section*{refname}%
% @mkboth{MakeUppercaserefname}{MakeUppercaserefname}%
list{@biblabel{@arabicc@enumiv}}%
{settowidthlabelwidth{@biblabel{#1}}%
leftmarginlabelwidth
advanceleftmarginlabelsep
@openbib@code
usecounter{enumiv}%
letp@enumiv@empty
renewcommandtheenumiv{@arabicc@enumiv}}%
sloppy
clubpenalty4000
@clubpenalty clubpenalty
widowpenalty4000%
sfcode`.@m}
{def@noitemerr
{@latex@warning{Empty `thebibliography' environment}}%
endlist}
makeatother
begin{document}
bibliographystyle{apacite}
noindent
Name : Goh Cheng Kee \
ID : 1121115581 \
Subject : TPT1201 Research Methods In Cs\
noindent
begin{longtable}{| p{.20textwidth} | p{.80textwidth} |}
hline
Paper Title & Dynamic Resource Allocation using Virtual
Machines for Cloud Computing Environment \
hline
Author(s) & Zhen Xiao, Senior Member, IEEE, Weijia Song, and Qi Chen \
hline
Problem Solved &
This research paper ~cite{xiao2013dynamic} study on how to allocate dynamic resources efficiently using virtual machines for cloud computing environment onto the physical hardware to solve under-utilized due to over-provisioned data centers for the peak demand. The algorithm they presented offering automatic scale up and down in response to load variation. It saves electricity as well as reducing the hardware cost.
These are the goal they aim to achieve:
begin{itemize}
item Overload avoidance: The capacity of the physical machines should fulfill the needs of Virtual Machines running on it. Otherwise, overloaded physical machines can decrease the performance of its virtual machines.
item Green computing: The number of active physical machines should be minimized as long as it can still satisfy the needs of all virtual machines. Inactive physical machines can be turned off to save energy.
end{itemize}
\
hline
Related work research &
On this section, we will discuss two main category of related work discussed on this paper.
begin{enumerate}[1.]%for capital roman numbers.
item Resource Allocation By Live VM
begin{itemize}
item VM live migration for dynamic allocation in a virtualized environment ~cite{wood2007black} ~cite{bobroff2007dynamic} ~cite{singh2008server} are methods uses for dynamic allocation and managing service-level agreement violations. It sort physical machines and the virtual machines based on their volumes when making the migration decision, however it cannot achieve green computing as it doesn't categories the cold spot and turn off the physical machines to save energy.
item On cite{singh2008server}, it mitigate not only the hot spots but also on network devices and storage nodes as well. The load balancing algorithm is a variant of the Toyoda ~cite{toyoda1975simplified} method for multi-dimensional knapsack problem.
end{itemize}
item Green computing
begin{itemize}
item On ~cite{nathuji2007virtualpower} it uses Dynamic Voltage and Frequency Scaling(DVFS) to adjust CPU power according to its load.
item On ~cite{nathuji2007virtualpower} it uses new hardware technologies such as Solid State Disk(SSD) and Self-Refresh DRAM to save energy. All green computing above are done in hardware as current research paper are done on virtual environment which reduce the cost of hardware.
end{itemize}
end{enumerate}
\
hline
Methodology &
In the paper, prediction are made based on the past behaviors of the virtual machine. In this case they monitor external behaviors instead of application level statistics because it requires modification on the Virtual Machines which always impossible to do so.
Using Exponentially Weighted Moving Average (EWMA) formula,
EWMA TCP-like scheme:
begin{equation}
E(t) = alpha * E(t – 1) + (1 – alpha) * O(t), 0 le alpha le 1
end{equation}
to measure the CPU load every minutes on the DNS server in their university and predict the load in the next minute. Although seemingly satisfactory, this formula does not capture the rising trends of resource usage. By setting $alpha$ to negative value, it resolve this problem. The formula can be transformed into following:
begin{equation}
begin{split}
E(t) = – | alpha | * E(t – 1) + (1 + | alpha |) * O(t) \
= O(t) + |alpha| * (O(t) – E(t – 1))
end{split}
end{equation}
On the other hand, the algorithm FUSD(Fast Up and Slow Down) is used to reduce the estimation when the observed resource usage is going down. Other prediction algorithm like Linear Auto-Regression models (SPAR) ~cite{spar1998} is used to models a predictive value as linear function of its past observations. Model parameters are determine by training with historical values. Comparing SPAR(4,2) and FUSD(-0.2,0.7), results in SPAR(4,2) achieves slightly better precision than FUSD(-0.2,0.7). However, the requirement of training phase to determine parameters is inconvenient, especially when the load pattern changes. Therefore they adopt the simpler EWMA variance.
end{longtable}
begin{longtable}{| p{.20textwidth} | p{.80textwidth} |}
& After predicted future resource demands of virtual machines and using skewness algorithm to quantify the unevenness in the utilization of multiple resources on a server. Servers are defines into three categories. Hot spot, Warm spot and cold spot. Hot spot servers is defined if the utilization of any of its resources is aboive a hot threshold (0.9). Warm spot is defined if the utilization that is sufficiently high to justify having the server running but not so high as to risk become a hot spot which is between hot spot and cold spot utilization values (0.65). Cold spot is defined if the utilization of all its resources are below a cold threshold(0.25). This indicates that the server is mostly idle and is potential servers to turn off to save energy. Green computing algorithm is invoked when the average utilization of all resources on active servers are below the green computing threshold (0.4). It have to migrate away its Virtual Machines before it can shut down an under-utilized server. For a cold spot servers, it have to migrate its Virtual Machines to somewhere else, the resource utilization of the server for accepting the migration of cold spot servers Virtual Machines must be below the warm threshold (0.65), but overdoing it might create hot spots in future. To solve this problem, accept a cold spot as a destination server if necessary. If the algorithm successfully find a destination servers for all virtual machines on a cold spot, the algorithm will record the sequence of migrations and update the predicted load of related servers.
The methodology they used above are effective in overload mitigation and green computing for every minute. But the prediction errors will increase when the CPU and memory loads changed dramatically in seconds, results in decreasing the performance of its virtual machines. Although, the SPAR(4,2) algorithm have better prediction compared to FUSD(-0.2,0.7), the prediction algorithm have to improve because it plays an important role in improving stability and performance.
\
hline
Claimed Contributions &
begin{itemize}
item Developed a resource allocation system that can avoid
overload in the system effectively while minimizing the
number of servers used.
item Introduced the concept of "skewness" to measure the
uneven utilization of a server and improved the overall utilization of servers in the multi-dimensional resource constraints.
item Designed a load prediction algorithm that can predict the
future resource usages require for a Virtual Machines without modifying the Virtual Machines to send its resource usage. The algorithm can capture the trend of resource usage patterns and help reduce the under-utilized servers significantly.
end{itemize}
\
hline
Conclusion &
The paper presented the design, implementation, and evaluation
of a dynamic resource allocation using virtual machines for cloud computing. Using skewness algorithm to combine virtual machines with different resource characteristics, so that the capacities of
servers are well utilized. In conclusion, the methodology presented on this paper achieves both overload avoidance and green computing.
\
hline
What did I learn. Future work &
In the end of this review, I learned the implementation of allocating Dynamic Resource using Virtual Machines for Cloud Computing Environment while minimizing under-utilized Physical Machines. I also learned different way of achieving Green Computing from the references cite by the authors. Besides, I learned how to conduct a research.
Future work may have done on this topic is the prediction algorithm. Minimize the median prediction errors and improve the provision time of Virtual Machines by implementing algorithms.
\
hline
References &
bibliography{MyBib}
\
hline
end{longtable}
end{document}