Next generation Cloud Computing Architecture

Enabling real-time dynamism for shared

distributed physical infrastructure
G.Monisha, M.Tech - !ear,"M.#aina Glancy, M.Tech - !ear,$M.%e&athi, M.Tech - !ear
'
A.(adi&elu,Assitant )rofessor,*epartment of nformation Technology
(elammal Engineering college,surapet,Chennai-++
Monigeorge,-.gmail.com, lainaglancy.gmail.com,sitams.&adi.&elu.gmail.com,hellore&athy.m.gmail.com
ABSTRACT
Cloud computing is fundamentally altering
the expectations for ho/ and /hen computing,
storage and net/or0ing resources should be
allocated, managed and consumed. End-users are
increasingly sensiti&e to the latency of ser&ices they
consume. 1er&ice *e&elopers /ant the 1er&ice
)ro&iders to ensure or pro&ide the capability to
dynamically allocate and manage resources in
response to changing demand patterns in real-time.
2ltimately, 1er&ice )ro&iders are under pressure to
architect their infrastructure to enable real-time
endto- end &isibility and dynamic resource
management /ith finegrained control to reduce total
cost of o/nership /hile also impro&ing agility. The
current approaches to enabling real-time, dynamic
infrastructure are inade3uate, expensi&e and not
scalable to support consumer mass-mar0et
re3uirements. 4&er time, the ser&er-centric
infrastructure management systems ha&e e&ol&ed to
become a complex tangle of layered systems
designed to automate systems administration
functions that are 0no/ledge and labor intensi&e.
This expensi&e and non-real time paradigm is ill
suited for a /orld /here customers are demanding
communication, collaboration and commerce at the
speed of light. Than0s to hard/are assisted
&irtuali5ation, and the resulting decoupling of
infrastructure and application management, it is no/
possible to pro&ide dynamic &isibility and control of
ser&ices management to meet the rapidly gro/ing
demand for cloud-based ser&ices. 6hat is needed is a
rethin0ing of the underlying operating system and
management infrastructure to accommodate the
ongoing transformation of the data center from the
traditional ser&er-centric architecture model to a
cloud or net/or0-centric model. This paper proposes
and describes a reference model for a net/or0-
centric datacenter infrastructure management stac0
that borro/s and applies 0ey concepts that ha&e
enabled dynamism, scalability, reliability and
security in the telecom industry, to the computing
industry.
Keywords-Cloud Computing, *istributed
Computing, (irtuali5ation, *ata Center
I. INTRODUCTION
The unpredictable demands of the 6eb ,.7 era in
combination /ith the desire to better utili5e T
resources are dri&ing the need for a more dynamic T
infrastructure that can respond to rapidly changing
re3uirements in real-time. This need for real-time
dynamism is about to fundamentally alter the
datacenter landscape and transform the T
infrastructure as /e 0no/ it.
n the cloud computing era, the computer can no
longer be thought of in terms of the physical
enclosure 8 i.e. the ser&er or box, /hich houses the
processor, memory, storage and associated
components that constitute the computer. nstead the
9computer: in the cloud ideally comprises a pool of
physical compute resources 8 i.e. processors,
memory, net/or0 band/idth and storage, potentially
distributed physically across ser&er and geographical
boundaries /hich can be organi5ed on demand into a
dynamic logical entity i.e. a 9cloud computer:, that
can gro/ or shrin0 in real-time in order to assure the
desired le&els of latency sensiti&ity, performance,
scalability, reliability and security to any application
that runs in it. 6hat is truly enabling this
transformation today is &irtuali5ation technology 8
more specifically hard/are assisted ser&er
&irtuali5ation. At a fundamental le&el, &irtuali5ation
technology enables the abstraction or decoupling of
the application payload from the underlying physical
resource ;,<. 6hat this typically means is that the
physical resource can then be car&ed up into logical
or &irtual resources as needed. This is 0no/n as
Proceedings of National Conference on Advanced Computing and CommunicationNCACC11, April.16, 2011
pro&isioning. =y introducing a suitable management
infrastructure on top of this &irtuali5ation
functionality, the pro&isioning of these logical
resources could be made dynamic i.e. the logical
resource could be made bigger or smaller in
accordance /ith demand. This is 0no/n as dynamic
pro&isioning. To enable a true 9cloud: computer,
e&ery single computing element or resource should
be capable of being dynamically pro&isioned and
managed in real-time. )resently, there are many
holes and areas for impro&ement in today>s
datacenter infrastructure before /e can achie&e the
abo&e &ision of a cloud computer. =elo/ /e discuss
these for each of the 0ey datacenter infrastructure
components.
A. Server Operating Systems and irt!a"i#ation
6hereas net/or0s and storage resources -
than0s to ad&ances in net/or0 ser&ices management
and 1ANs, ha&e already been capable of being
&irtuali5ed for a /hile, only no/ /ith the /ider
adoption of ser&er &irtuali5ation do /e ha&e the
complete basic foundation for cloud computing i.e.
all computing resources can no/ be &irtuali5ed.
Conse3uently, ser&er &irtuali5ation is the spar0 that is
no/ dri&ing the transformation of the T
infrastructure from the traditional ser&er centric
computing architecture to a net/or0-centric, cloud
computing architecture. 6ith ser&er &irtuali5ation,
/e no/ ha&e the ability to create complete logical
?&irtual@ ser&ers that are independent of the
underlying physical infrastructure or their physical
location. 6e can specify the computing, net/or0 and
storage resources for each logical ser&er ?&irtual
machine@ and e&en mo&e /or0loads from one &irtual
machine to another in real-time ?li&e migration@. All
of this has helped to radically transform the cost
structure and efficiency of the data center. Capacity
utili5ation of ser&ers can be increased and o&erall
consolidating /or0loads can dramatically reduce
po/er consumption. Additionally, than0s to ser&er
&irtuali5ation and li&e migration, Aigh A&ailability
?AA@ and *isaster %eco&ery ?*%@ can be
implemented much more efficiently ;B<. *espite the
numerous benefits that &irtuali5ation has enabled /e
are yet to reali5e the full potential of &irtuali5ation in
terms of cloud computing. This is becauseC
Traditional server-centric operating sstems !ere
not designed to manage s"ared distri#uted
resources$
The Cloud computing paradigm is all about
optimally sharing a set of distributed computing
resources /hereas the ser&er-centric computing
paradigm is about dedicating resources to a particular
application. The ser&er-centric paradigm of
computing inherently ties the application to the
ser&er. The Dob of the ser&er operating system is to
dedicate and ensure a&ailability of all a&ailable
computing resources on the ser&er to the application.
f another application is installed on the same ser&er,
the operating system /ill once again manage all of
the ser&er resources, to ensure that each application
continues to be ser&iced as if it has access to all
a&ailable resources on that ser&er. This model /as
not designed to allo/ for the 9dial-up: or 9dialdo/n:
of resource allocated to an application in response to
changing /or0load demands or business priorities.
This is /hy load-balancing and clustering /as
introduced. Ao/e&er, that does not alter the
association of an application to a ser&er. t Dust uses
more instances of the application 8 each running in
their o/n ser&er, to try and share any increased
burden. 6hat is re3uired for cloud computing 8
/here distributed resources are shared amongst
applications, is for a /ay to 9mediate: bet/een the
applications and the resources by prioriti5ing the
applications> needs based on relati&e business
priorities. 4ur 0ey obser&ation here is that any
sharing of resources /ill at some point ine&itably
result in contention for those resources /hich can
only be resol&ed through a system that performs
mediation globally across all the distributed shared
resources. Today>s operating systems do not nati&ely
pro&ide this type of capability. This is often relegated
to management systems that are layered on top or
orthogonal to operating systems. Ao/e&er, the
management system /ere also designed for a ser&er-
centric, configuration based paradigm and ha&e
similar issues /hich ma0e them ill suited as
mediators that can enable real-time dynamism. The
issues related to management systems are detailed in
a separate section belo/.
Current "pervisors do not provide ade%uate
separation #et!een application management and
p"sical resource management$
Today>s hyper&isors ha&e Dust interposed
themsel&es one le&el do/n belo/ the operating
system to enable multiple 9&irtual: ser&ers to be
hosted on one physical ser&er. 6hile this is great for
consolidation, once again there is no /ay for
applications to manage ho/, /hat and /hen
resources are allocated to themsel&es /ithout ha&ing
to /orry about the management of physical
resources. t is our obser&ation that the current
generation of hyper&isors /hich /ere also born from
the era of ser&er-centric computing does not
delineate hard/are management from application
management much li0e the ser&er operating systems
themsel&es. t is our contention that management and
allocation of a shared infrastructure re3uire a
different approach.
B. Storage Networ$s % irt!a"i#ation
=efore the proliferation of ser&er &irtuali5ation,
storage net/or0ing and storage &irtuali5ation enabled
many impro&ements in the datacenter. The 0ey dri&er
/as the introduction of the Eibre Channel ?EC@
protocol and Eibre Channel-based 1torage Area
Net/or0s ?1AN@ /hich pro&ided high speed storage
connecti&ity and speciali5ed storage solutions to
enable such benefits as ser&er-less bac0up, point to
point replication, AAF*% and performance
optimi5ation outside of the ser&ers that run
applications. Ao/e&er, these benefits ha&e come
24
Ad"iparasa&t"i 'ngineering College, (elmaruvat"ur
/ith increased management complexity and costs. n
fact 1AN administrator costs are often cited as the
single most critical factor affecting the successful
deployment and management of &irtual ser&er
infrastructure.
&. Networ$ irt!a"i#ation
The &irtual net/or0s no/ implemented inside the
physical ser&er to s/itch bet/een all the &irtual
ser&ers pro&ide an alternati&e to the multiplexed,
multi-pathed net/or0 channels by trun0ing them
directly to 6AN transport thereby simplifying the
physical net/or0 infrastructure. 6ith the
proliferation of multi-core multi-C)2 commodity
ser&ers, it has almost become necessary to eliminate
the mess of cables other/ise needed to interface
multiple A=As and NCs for each application /ith a
single high speed Ethernet connection and a &irtual
s/itch. t is our contention that resultant architectural
simplicity /ill significantly reduce associated
management burden and costs.)resent day
management systems are not cut out to enable the
real-time dynamic infrastructure needed for cloud
computing. Aere are the reasons /hyC
)uman sstem administrators do not lend
t"emselves to ena#ling real-time dnamism$
Polic-#ased management is not reall automation$
There are a lot of management systems
today that pro&ide policy-based management
capabilities. The trouble is the policies ha&e to be
programmed by expert system administrators /ho
ma0e Dudgments based on their experience. This is
neither an optimal nor a scalable solution for cloud
computing en&ironments /here the /or0load
demands are unprecedented and &ary /ildly.
'. App"i&ation Creation and (a&$aging)
The current method of using (irtual
Machine images that include the application, 41 and
storage dis0 images is once again born of a ser&er-
centric computing paradigm and does not lend itself
to enable distribution across shared resources. n a
cloud computing paradigm, applications should
ideally be constructed as a collection of ser&ices
/hich can be composed, decomposed and distributed
on the fly. Each of the ser&ices could be considered
to be indi&idual processes of a larger /or0flo/ that
constitutes the application. n this /ay, indi&idual
ser&ices can be orchestrated and pro&isioned to
optimi5e the o&erall performance and latency
re3uirements for the application.
II. (RO(OS'D R'*'R'NC'
ARC+IT'CTUR' ,OD'-
f /e /ere to distill the abo&e obser&ations
from the pre&ious section, /e can see a couple of 0ey
themes emerging.
That isC
.. The next generation architecture for cloud
computing must completely decouple physical
resources management from &irtual resource
managementG and
/. )ro&ide the capability to mediate bet/een
applications and resources in real-time. As /e
highlighted in the pre&ious section, /e are yet to
achie&e perfect decoupling of physical resources
management from &irtual resource management but
the introduction and increased adoption of hard/are
assisted &irtuali5ation ?AA(@ as an important and
necessary step to/ards this goal. Than0s to AA(, a
next generation hyper&isor /ill be able to manage
and truly ensure the same le&el of access to the
underlying physical resources. Additionally, this
hyper&isor should be capable of managing both the
resources located locally /ithin a ser&er as /ell as
any resources in other ser&ers that may be located
else/here physically and connected by a net/or0.
4nce the management of physical resources is
decoupled from the &irtual resource management the
need for a mediation layer that arbitrates the
allocation of resources bet/een multiple applications
and the shared distributed physical resources
becomes apparent.
0. *nfrastructure +ervice ,a#ric$
This layer comprises t/o pieces. Together
the t/o components enable a computing resource
9dial-tone: that pro&ides the basis for pro&isioning
resource e3uitably to all applications in the cloudC
.. -istri#uted +ervices (ediation$ This is a
ECA)1based ?Eault, Configuration, Accounting,
)erformance and 1ecurity@ abstraction layer that
enables autonomous self-management of e&ery
indi&idual resource in a net/or0 of resources that
may be distributed geographically, and a
,. Virtual Resource Mediation Layer: This pro&ides
the ability to compose logical &irtual ser&ers /ith a
le&el of ser&ice assurance that guarantees resources
such as number of C)2s, memory, band/idth,
latency, 4)1 ?F4 operations per second@, storage
throughput and capacity.
-istri#uted +ervices Assurance Platform$ This layer
/ill allo/ for creation of ECA)1-managed &irtual
ser&ers that load and host the desired choice of 41 to
allo/ the loading and execution of applications.
1ince the &irtual ser&ers implement ECA)1-
management, they can pro&ide automated mediation
ser&ices to nati&ely ensure fault management and
reliability ?AAF*%@, performance optimi5ation,
accounting and security. This defines the
management dial-tone in our reference architecture
model. 6e en&ision that ser&ice pro&iders /ill offer
these &irtual ser&ers /ith appropriate management
A) ?management dial-tone@ to the ser&ice
25
Proceedings of National Conference on Advanced Computing and CommunicationNCACC11, April.16, 2011
de&elopers to create self-configuring, self-healing,
self optimi5ing ser&ices that can be composed to
create self-managed business /or0flo/s that are
independent of the physical infrastructure.
III. D'(-O1,'NT O* T+' R'*'R'NC'
,OD'-
Any generic cloud ser&ice platform
re3uirements must address the needs of four
categories of sta0e holders ?H@ nfrastructure
)ro&iders, ?,@ 1er&ice )ro&iders. ?B@ 1er&ice
*e&elopers, and ?'@ End 2sers. =elo/ /e describe
ho/ the reference model /e described /ill affect,
benefit and be deployed by each of the abo&e
sta0eholders.
*nfrastructure providers: These are &endors /ho
pro&ide the underlying computing, net/or0 and
storage resources that can be car&ed up into logical
cloud computers, /hich /ill be dynamically
controlled to deli&er massi&ely scalable, and globally
interoperable ser&ice net/or0 infrastructure. =oth
ser&ice creators /ho de&elop the ser&ices and also
the end users /ho utili5e these ser&ices /ill use the
infrastructure. This is &ery similar to s/itching,
transmission and access e3uipment &endors in the
telecom /orld /ho incorporate ser&ice enabling
features and management interfaces in their
e3uipment. Current storage and computing ser&er
infrastructure has neither the ability to dynamic dial-
up and dial-do/n resources nor the capability for
dynamic management /hich /ill help eliminate the
numerous layers of present day management systems
and the human latency they contribute. The ne/
reference architecture pro&ides an opportunity for the
infrastructure &endors to eliminate current systems
administration oriented management paradigm and
enable next generation real-time, on-demand,
ECA)1-based management so that applications can
dynamically re3uest the dial-up and dial-do/n of
allocated resources.
+ervice providers: 6ith the deployment of our ne/
reference architecture, ser&ice pro&iders /ill be able
to assure both ser&ice de&elopers and ser&ice users
that resources /ill be a&ailable on demand. They /ill
be able to effecti&ely measureand meter resource
utili5ation end-to-end usage to enable a dial-tone for
computing ser&ice /hile managing 1er&ice #e&els to
meet the a&ailability, performance and security
re3uirements for each ser&ice. The ser&ice pro&ider
/ill no/ manage the application>s connection to
computing, net/or0 and storage resource /ith
appropriate 1#As. This is different from most current
cloud computing solutions that are nothing more than
hosted infrastructure or applications accessed o&er
the nternet. This /ill also enable a ne/ distributed
&irtual ser&ices operating system that pro&ides
distributed ECA)1-based resource management on
demand.
+ervice -evelopers$ They /ill be able to de&elop
cloudbased ser&ices using the management ser&ices
A) to configure, monitor and manage ser&ice
resource allocation, a&ailability, utili5ation,
performance and security of their applications in
real-time. 'nd .sers$ Their demand for choice,
mobility and interacti&ity /ith intuiti&e user
interfaces /ill continue to gro/. The managed
resources in our reference architecture /ill no/ not
only allo/ the ser&ice de&elopers to create and
deli&er ser&ices using logical ser&ers that end users
can dynamically pro&ision in real-time to respond to
changing demands, but also pro&ide ser&ice
pro&iders the capability to charge the end-user by
metering exact resource usage for the desired 1#A.
. CONC-USION
n this paper, /e ha&e described the
re3uirements for implementing a truly dynamic
cloud-computing infrastructure. 1uch an
infrastructure comprises a pool of physical
computing resources 8 i.e. processors, memory,
net/or0 band/idth and storage, potentially
distributed physically across ser&er and geographical
boundaries /hich can be organi5ed on demand into a
dynamic logical entity i.e. 9cloud computer:, that can
gro/ or shrin0 in real-time in order to assure the
desired le&els of latency sensiti&ity, performance,
scalability, reliability and security to any application
that runs in it. 6e identified some 0ey areas of
deficiency /ith current &irtuali5ation and
management technologies. n particular /e detailed
the importance of separating physical resource
management from &irtual resource management and
/hy current operating systems and hyper&isors 8
/hich /ere born of the ser&er-computing era, are not
designed and hence illustrated to pro&ide this
capability for the distributed shared resources typical
of cloud deployment. 6e also highlighted the need
for ECA)1 based ?Eault, Configuration, Accounting,
)erformance and 1ecurity@ ser&ice 9mediation: to
pro&ide global management functionality for all
net/or0ed physical resources that comprise a cloud 8
irrespecti&e of their distribution across many physical
ser&ers in different geographical locations. 6e then
proposed a reference architecture model for a
distributed cloud computing mediation
?management@ platform, /hich /ill form the basis
for enabling next generation cloud computing
infrastructure. 6e sho/ed ho/ this infrastructure
/ould affect as /ell as benefit 0ey sta0eholders such
as the nfrastructure pro&iders, ser&ice pro&iders,
ser&ice de&eloper and end-users.
R'*'R'NC'S
;H< %ao Mi00ilineni, (iDay 1arathy ICloud Computing and
#essons from the )astI, )roceedings of EEE 6ETCE ,77J, Eirst
nternational 6or0shop on Collaboration K Cloud Computing,
Lune ,77J
;,< %aD0umar =uyyaa, Chee 1hin !eoa, , 1ri0umar (enugopala,
Lames =roberga, and &ona =randicc, 9Cloud computing and
emerging T platformsC (ision, hype, and reality for deli&ering
computing as the M
th
utility:, Euture Generation Computer
1ystems, (olume ,M, ssue +, Lune ,77J, )ages MJJ-+H+
;B< Lason A. Nappel, Anthony T. (elte, Toby L. 6elte, 9Microsoft
(irtuali5ation /ith Ayper-(:, McGra/ Aill, Ne/ !or0, ,77J
26
Ad"iparasa&t"i 'ngineering College, (elmaruvat"ur
;'< *a&id Chisnall, 9guide to the xen hyper&isor:, Eirst edition,
)rentice Aall, )ress, NL, ,77J
;M< Gartner>s ,77O *ata Center Conference nstant )olling
%esultsC (irtuali5ation 1ummary 8 March ,, ,77J.
27