virtu l - dlt solutions
TRANSCRIPT
04 Public vs. Private:ComparingCloudArchitecturesBefore picking either option, it’s important to examinethe practical limitations of private and public clouds.BY STEPHEN J. BIGELOW
12 How toSet PrioritiesforVirtualizingApplicationsDon't forget about user experience when establishingyour virtual application plan. BY MIKE NELSON
17 Measuring andOptimizing theCloudEvaluate performance data in a consistent mannerwith tools specially designed for the cloud. BY BRIEN M. POSEY
virtu lDATA CENTER
v
Volume 32
2 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
RE YOU in the cloud? Arecent TechTarget sur-vey said that more than15% of respondentsare using some type
of cloud, while 17% plan to implementthe technology in the next 12 months.Before taking the plunge, find out
which cloud computing architecture isbest for your organization. And per-haps more important, learn the draw-backs of each option to make surethey aren’t deal-breakers.For example, organizations that opt
for private clouds can look forward tomaintaining their investment in theirexisting infrastructures. But privateclouds may raise compliance issuesbecause it’s difficult to track the phys-ical location of data.And those choosing a public cloud
might be enticed by the cost savingsbecause they don’t have to buy, in-stall, operate or maintain servers orother equipment. At the same time,though, organizations working with acloud provider will have to deal withtradeoffs in reliability and control. Digdeep into the differences in Stephen J.Bigelow’s “Public vs. Private: Compar-ing Cloud Architectures.”
In the cloud, it’s important to re-member that your users don’t knowand probably don’t care where anapplication is running. They expectthe application to perform just aswell as it did when it was hosted inyour data center. So how can youmake sure that your applications areperforming adequately in a cloudenvironment? Read Brien M. Posey’s“Measuring and Optimizing theCloud” for a new focus on cloudperformance.There are no hard-and-fast rules
for what you can or should virtualizein your environment. In some cases,you might think twice about virtualiz-ing an application if it would no longerlook the same on the users’ end. MikeNelson categorizes the applicationsthat can be hosted in a virtual envi-ronment in “How to Set Priorities forVirtualizing Applications.”How are you taking your IT opera-
tions to the next level? Send me anemail at [email protected],and tell me all about it. �
CHRISTINE CASATELLI
Editor, Virtual Data Center
WeighingYourOptions
A
CLOUD NETWORKING NOWHAS AN ETHERNET FABRIC
PUT THE CLOUD WITHIN REACH.
A cloud-optimized network may be closer than you think. With revolutionary
Brocade® VDX™ 6720 Data Center Switches featuring Brocade VCS™ technology,
you get the industry’s first true Ethernet fabric solution that can help dramatically
simplify your architecture and streamline management. The result is unequalled
network performance, greater utilization—and a clear path to the cloud.
© 2011 Brocade Communications Systems, Inc. All Rights Reserved. Brocade and the B-wing symbol are registered trademarks of Brocade Communications Systems, Inc. All other brands, products, or service names mentioned are or may be trademarks or service marks of their respective owners.
LEARN MORE AT www.brocade.com/vdx
4 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
T’S NO secret that cloud computinghas been a misunderstood victimof marketing hype. But regardlessof how you define the technolo-gies, it’s clear that both private
and public cloud computing are slowlybeing adopted among modern busi-nesses. According to a recent Tech-Target survey, roughly 15% of morethan 1,000 IT respondents indicatedthat they have implemented sometype of cloud, while almost 17%expect to implement some type ofcloud computing over the next 12months.The successful move to a cloud is
hardly quick or simple, and perhapsthe biggest challenge that IT adminis-trators face is choosing between pri-vate and public cloud computingarchitectures. Before picking eitheroption, it’s important to examine thepractical limitations of private andpublic cloud computing.First, let’s consider the private
cloud, which may be a dedicated, orleased, portion of a provider’s hostedfacilities. But in most cases, a privatecloud is the conversion of an existingtraditional data center into a utility-based computing architecture—youown it, secure it and maintain it justas you always have.The main difference is that a pri-
vate cloud shifts the emphasis tocomputing on demand. This generallychanges the way that data centerhardware is used and shifts the waythat departments or stakeholders payfor those computing resources.There are many reasons for organi-
zations to select private clouds. Costis always an obvious issue, but it’s notnecessarily the principal motivator.Organizations that opt for privateclouds are usually most concernedwith preserving their investment intheir existing infrastructures. This wasthe reason cited by more than 35% ofTechTarget survey respondents, and it
Public vs. Private:ComparingCloudArchitecturesBEFORE PICKING EITHEROPTION, IT’S IMPORTANTTO EXAMINETHE PRACTICAL LIMITATIONSOF PRIVATE ANDPUBLIC CLOUDS.BYSTEPHEN J. BIGELOW
I
5 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
makes sense for businesses that havealready spent significant capital tocreate their data centers in the firstplace.Another 17% said they were moti-
vated by disaster recovery (DR) andbusiness continuity benefits, while16% said they wanted the security ofcontinued IT ownership. Eleven per-cent said they sought to leverage theself-service and automation capabili-ties that are a hallmark of true cloudcomputing.
PRIVATE CLOUDS AND COMPLIANCEData location may also become acompliance or regulatory issue forprivate clouds if the organizationdeploys them across geographicboundaries. It’s extremely difficult totrack the physical location of data ina true cloud—as opposed to a virtual-ized data center—so close attentionto regulatory adherence is stillrequired.Even when an organization deploys
a private cloud successfully, makingthe most of that cloud can be a formi-dable task. Efficient cloud manage-ment involves the establishment ofstandard operating procedures, a highlevel of automation, sharing of com-puting resources among departmentsor stakeholders, along with a regi-mented standardization through a sin-gle service-level agreement (SLA).This set of criteria allows an admin-
istrator to run the entire pool of cloudresources at an exceptionally high uti-lization rate and take advantage ofsignificant cost benefits. But these cri-
teria are not normal considerations ina typical data center. “If you don’thave the discipline of the regimentedresources, if you don’t have automa-tion to drive this forward, and if youdon’t have sharing, you can’t reachthat degree of utilization,” said JamesStaten, vice president and principalanalyst at Forrester Research.
NOT AN ALL-OR-NOTHING DECISIONThe transition to a private cloud isnot an all-or-nothing decision. Ex-perts recommend a gradual transition,starting with a limited cloud infra-structure and running new or low-pri-ority applications on that “mini cloud”within the environment. This cautiousapproach allows ample time to gainfamiliarity and comfort with cloudtechnology and examine the impactof tradeoffs on your particular organi-zation. Success may accelerate subse-quent projects, while problems mayslow further transition. But a well test-ed alpha phase of adoption can mini-mize unnecessary disruption to yourbusiness.Data protection within the private
cloud is another important considera-tion. Organizations that prefer moretraditional backup mechanisms cancontinue to use that approach in thecloud, but the backup tools—such asSymantec's Backup Exec—must be up-graded to a cloud-compatible version.Other organizations with more
exacting recovery needs may optfor some combination of replication/snapshots and clustering as a meansof mixing high availability and data
6 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
protection for fast restoration. Perhapsthe biggest challenge is disasterrecovery (DR). The notion of nativeDR in a cloud is a misconception, andadopters must pay close attention tothe DR plan implemented in their pri-vate cloud.Deploying and testing data protec-
tion schemes is another central rea-son why companies should approachprivate cloud adoption gradually. Thisapproach will leave mission-critical
applications with the most stringentdata protection requirements until theprivate cloud is well understood anddeveloped.A public cloud employs the same
basic technologies used to create aprivate cloud, but public clouds suchas Microsoft’s Azure serve a funda-mentally different purpose. In virtuallyall cases, a public cloud is accessedas a service through a provider thathosts all of the hardware, services and
FOR A PUBLIC cloud to be economical, it’s imperative for the provider to share com-puting resources among different users and automate every possible aspect of thebusiness. The net result is an inexpensive—yet inflexible—source of utility com-puting.
True public cloud providers—as opposed to hosted service providers that adver-tise themselves as cloud providers—just cannot afford to cater to the uniqueneeds of individual clients. As a result, service-level agreements (SLAs) are quiterigid and almost always side with the provider, especially when it comes to securi-ty and data protection.
The notion of a business committing its important applications and data to thecare of a cloud provider that assumes no significant responsibility for the safety,availability or even the physical location of that data is enough to scare off eventhe hardiest CTOs. But experts say that this is slowly changing and will continue toshift as public cloud providers develop the tools and tactics necessary to addressserious business concerns.
Changes are coming to public cloud SLAs, said Laurie McCabe, partner with theSMB Group in Northboro, Mass. But McCabe said she sees the emphasis squarelyon compensation for downtime, which can have the most immediate impact onuser productivity.
“I’m losing money because I can’t enter sales or whatever it is,” McCabe said.“That’s where cloud providers will focus their efforts,” she said. At the moment,there is no generally accepted formula for downtime compensation, but it’s an areathat public cloud providers are targeting, McCabe said. �
ANOTHER LOOKATSLAs
7 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
applications. Public cloud operationalso leverages significant economiesof scale, potentially supporting vastserver and storage resources spreadacross numerous geographicalareas—even globally.
COST SAVINGS WITHPUBLIC CLOUDSThe most prevalent benefit in a publiccloud is cost savings. Organizationsdo not buy, install, operate or main-tain servers or other equipment asthey would with private clouds.Organizations may develop their ownapplications, but they are hosted by apublic cloud provider.Public clouds are also designed
as true multi-tenant environments.This allows a huge number of usersto share a provider’s computing re-sources, which also makes publiccloud services highly cost-efficient forusers. Experts agree that the cost of apublic cloud is typically far lower thana traditional data center and privatecloud because there is no capitalexpense. There is also greater flexibili-ty for users because organizations payonly for the computing resources thatthey actually use.According to the respondents who
use a public cloud, more than 31%said they were motivated by a reduc-tion in infrastructure costs, another26% sought more flexibility than tra-ditional hosted services, almost 21%wanted a more scalable infrastructurewithout up-front costs, and 11%sought lower IT staff costs.Security, reliability and regulatory
compliance remain the principal con-cerns with public clouds. The conceptof multi-tenancy supports strong eco-nomics by allowing high utilization.There is persistent concern about po-tential vulnerabilities or attacks thatmight be used to exploit the hypervi-sor and affect other users’ sessions.Although there have been no reportsof successful attacks, security re-mains a serious point of contentionfor users considering public clouds.“Security is a shared burden but
not shared equally,” said Staten, add-
ing that services like Azure providesecurity to the point of abstractionbut no higher. For example, Azurewould presumably be responsiblefor server security by providing logsto ensure that no violations occurred.But organizations are responsible
for security on their own applications.That includes design oversights likeopen ports or weak logins. “This un-even handshake is the mistake mostpeople make when they use cloud
Although there havebeen no reports ofsuccessful attacks,security remains aserious point of con-tention for users con-sidering public clouds.
8 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
resources in an unsecure fashion,”Staten said.Reliability and availability are other
issues in public clouds. Even whenproviders implement clusters, redun-dant sites and other tactics to im-prove availability, outages can occurperiodically. Because downtime canaffect the ability to access cloudapplications—and affect your organi-zation’s ability to function—the move
to a public cloud should include acareful evaluation of availability andthe consequences of normal andextended downtime.
LOCATION, LOCATION, LOCATIONThe physical location of sensitive datamay also become a regulatory prob-lem for public clouds. For example,some industries are not allowed to
FOR BUSINESSES TO leverage public cloud technology, data must be secure andaccessible to users. Today’s global challenges make this goal far from certain.
Companies do encounter natural and man-made disasters, fall victim to hackingand go out of business entirely. What happens if your cloud provider falters?
Experts say that any move to the public cloud should include a careful consider-ation of application and data portability. For example, it may be necessary to moveaway from one cloud provider and engage another. Otherwise, you might need torestore that application within your own data center—effectively taking the appli-cation out of the cloud.
In either case, part of cloud disaster preparation involves some form of transi-tion plan. “The role of IT is changing,” said Laurie McCabe, partner with the SMBGroup in Northboro, Mass. “IT people have to have a strategy or plan for determin-ing what they are comfortable with, creating a contingency plan should they needto make an adjustment.”
Some businesses develop a relationship with a second public cloud provider andhave already worked out a process for mapping their applications to the secondenvironment. Other organizations may adopt gateway tools such as CloudSwitchto minimize architectural changes and ease application porting between clouds.And still other organizations may retain and periodically test local servers that canstart up virtual machines to run applications in-house.
The good news is that most businesses have time to think ahead and make plansfor a transition away from or between clouds. As long as steps are taken beforetrouble actually strikes, planning for a serious transition is simply another aspectof disaster planning. �
PREPARING FORDISASTER
store data outside certain geographi-cal boundaries, but a true cloud doesnot observe such boundaries.A file created by a company in
Cleveland may wind up on a cloudprovider’s storage array in Toronto or
London. If an organization cannot pro-vide the physical location of its data,it’s impossible to pass a regulatoryaudit, and this may expose an organi-zation to serious legal issues. Ulti-mately, it’s important to examine theprovider’s SLA closely and understandthe level of security, reliability andregulatory support.As with the private cloud, a transi-
tion to a public cloud should be madeslowly and methodically, starting withlow-priority applications and services.In many cases, these are new applica-tions and services that are not even indata centers yet. For example, anorganization may continue using anexisting tool in-house but adopt a newcloud version to see how the two ver-
sions compare or perform.The point here is that a true public
cloud—as opposed to a hosted serv-ice that’s simply labeled as a cloud—is a one-trick pony with seriousrestrictions that may render it inade-quate for every application or service.“When you go to the self-service por-tal where you deploy your application,you’re confronted by all of theserestrictions,” Staten said. “If it’s a truepublic cloud, then you have to adaptto it—it doesn’t adapt to you.”Don’t assume that a public cloud
will protect your data. Although acloud provider can rebuild lost con-tent to some extent, it may be farbelow the protection that your appli-cation and its data require. For exam-ple, it may be necessary to back updata from the cloud to a local storagesystem and then replicate that datayourself for DR purposes. A risk an-alysis can help identify the level ofprotection needed for your data andallow you to plan for that protection.
HYPE VS. REALITYDeciding between a public and pri-vate cloud can be a serious challengebecause each one is not appropriatefor every application, user group ororganization. The hype that surroundscloud computing has caused confu-sion, prompting many organizationsto steer clear of the technologyentirely.According to TechTarget’s recent
survey, more than 40% of IT respon-dents said they have chosen to avoidcloud computing because of security
9 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
Aswith the privatecloud, a transition toa public cloud shouldbemade slowly andmethodically, startingwith low-priorityapplications andservices.
concerns. Another 30% of respon-dents said that their business applica-tions do not translate well to a cloudenvironment, while 27% said thatthey do not meter or charge back forcomputing resource use. More than26% of respondents expressed con-cern about data storage integrity andcompliance, and the remainingrespondents cite concerns over virtu-alization, in-house skills and cloudcomplexity—among others—as rea-sons for not using cloud technology.But experts warn organizations
that do venture into cloud technologynot to underestimate the challenges.Cloud computing provides enormouspotential for efficiency and economy,but achieving those goals within acloud is difficult. “If you don’t put theright kind of workload into a cloudenvironment, it will not yield cost sav-ings,” Staten said. “In fact, it may costyou more.”Generally speaking, the best appli-
cations for a cloud deployment areWeb-based, customer-facing appli-cations where many users need toaccess applications from various loca-tions using different levels of connec-tivity. Examples often include email,collaboration and support portals.These are the low-hanging fruit ofcloud computing.The case for a cloud is not so
easy for applications that are heavilytweaked or used only by a small groupat one location. These may be betterserved with a more traditional client/server deployment in an organiza-tion’s data center.
EFFICIENCY OF CLOUD ECONOMICSBy its very nature, an efficient cloudis a highly automated and regimentedenvironment that can provide only afixed set of assets and features tousers, Staten said. When organiza-tions try to satisfy the unique needsand wants of every stakeholder, theefficiency of cloud economics isquickly lost.Not only are assets and features
limited in the cloud, but applicationsare also aggressively removed so thatcomputing resources can be reused asneeded. This concept may be hard tograsp for many organizations whereapplications seem to consume com-puting resources long after their use-fulness is over. “This is a concept thatis hard for a lot of enterprises to get,which is why we see so few true pri-vate clouds,” Staten said.Ultimately, experts see many of
the biggest cloud concerns easing inthe coming years. For example, cloudvendors are actively working to over-come data location worries. As aresult, some form of accreditationmay eventually identify cloudproviders that adopt verifiable datalocation and protection practices.The future may yield a broader range
of clouds, each providing differentsets of assets and catering to moredemanding user requirements, Statensaid. At the same time, it’s importantto remember that neither public norprivate clouds will handle every com-puting need. Traditional data centerswill continue well into the future—even as the clouds move in. �
10 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
©20
11 C
isco
Sys
tem
s, In
c. A
ll rig
hts
rese
rved
.move fearlesslyamong the clouds.it’s open, yet secure. agile, yet efficient. virtual, yet real.
it builds on your current network, combining servers, networking, storage and virtualization. so you can deploy applications in minutes, not weeks.
move among the clouds. grow your business. and turn silos into new ways of working together.
introducing the Cisco data center business advantage.
cisco.com/go/dcbusiness
12 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
ECIDINGWHICHapplica-tions to run on virtualservers can be diffi-cult. Virtualizingapplications requires
careful consideration, so it’s best toprioritize which apps are the best can-didates. Successfully virtualized appli-cations improve resource usage anddisaster recovery, reduce hardwareand power costs and provide greaterflexibility.IT administrators, management and
business people alike often ask methe same question: “Which applica-tions do you recommend moving tovirtual servers?” There are no hard-and-fast rules for what you can orshould virtualize in your environment.The benefits of server consolidationand virtualization have motivatedmany organizations to run virtualizedapplications, even mission-criticalor resource-intensive apps.But establishing your priorities for
virtualizing applications is not as
daunting as you might think. Theapplications that can be hosted in avirtual environment can be dividedinto four buckets. The following aremy unconventional names for thosecategories:
� Heavy hitters. This categoryincludes tier-one applications—andapplications that have consistentlyhigh resource utilization. That in-cludes some SAP programs, financialapps and highly transactional data-bases. This category also includesany mission-critical redundancy-candidate applications.
� Hills and valleys.Most of thetime, these applications use mediumto high levels of resources, but theyalso have slow periods of barely anyresource usage—although these peri-ods do not last long. In this category,month-end financial and sales appli-cations are my favorite candidatesfor virtualized applications.
HowtoSetPrioritiesforVirtualizingApplicationsDON'TFORGETABOUTUSER EXPERIENCEWHEN ESTABLISHINGYOURVIRTUALAPPLICATION PLAN. BYMIKE NELSON
D
13 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
� Smooth and steady.Generally,these applications have steadyresource usage and experience usagespikes only briefly—if at all. For exam-ple, a sudden need for more resourcesmay coincide with events such asbackups or antivirus scans.
� Hardware lockers. Very few exist,but some older applications stilldemand hardware dependencies forlicense compliance or direct-consoleaccess. Most of these application ver-sions are no longer produced or evensupported, but for one reason oranother, some shops just cannot partwith them. Most of these apps cannotor should not function as virtualizedapplications.
If getting the most immediatebang for your buck is your goal, I sug-gest virtualizing applications in thesmooth-and-steady category. Consid-er license servers, monitoring applica-tions, patching services and evenmost file and print applications. Justabout any application that has lower-end utilization in any combination ofthe four core resources—RAM, CPU,disk and network—is a good candi-date for virtualization.Virtualized applications in the top
two categories—the heavy-hitters andthe hills and valleys—require moreresources, testing and legwork. SomeIT pros say you should not or cannotvirtualize these types of applications,but I don’t believe in blanket state-ments. Even mission-critical applica-tions are possibilities for virtualiza-tion. Still, some common virtualization
roadblocks include vendor supportand licensing issues. It’s best to deter-mine support and licensing rulesbefore you invest a lot of time andenergy in your plan to virtualize appli-cations.
Remember to cover all your bases.Run performance testing and base-lines on a physical server, then com-pare the results to those on a test vir-tual server. Make sure that you keepthe application owners or usersinformed about the process. Andmost of all, don’t be afraid to pushback on application vendors. Theysometimes offer the most resistanceto virtualizing applications.Once you prioritize which applica-
tions to virtualize, it’s a good idea totest their performance. Applicationperformance testing is the best way tolearn how virtual applications will farein your virtual infrastructure and what
Just about anyapplication that haslower-end utilizationin any combinationof the four coreresources—RAM,CPU, disk and network—is a good candidatefor virtualization
14 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
the end-user experience will be like.Virtual application performance
testing is exactly that—testing. Somepeople consider it a science, but Ithink of it as an art form. You can getcreative with testing to generate thebest performance, but it won’t exactlyreplicate how your virtual applicationswill perform in production.
PRE-TESTING FACTORS:SOFTWARE AND USERSApplication performance testing isan essential part of any solid virtualapplication plan, particularly for tier-one and mission-critical applications.By performing some quick load tests—with some willing users enlisted asguinea pigs—you can experiment withthe hypervisor and the apps them-selves.But as you tinker with virtual appli-
cations, don’t forget about user expe-rience. If a virtualized application runsas well as it did on physical hard-ware—or better—you shouldn’t worrytoo much about user expectations.In some cases, though, you might
think twice about virtualizing an appli-cation if, as a result, it would no longerlook the same on the users’ end. Forexample, if an application is visuallyintensive and is not a true client/server app, but rather a simpleRemote Desktop Protocol session-accessed application, a user’s dailyinteractions with the application aremost important. Virtualizing applica-tions saves resources and space—which is great for admins—but itwon’t always benefit users.
Finally, remember that not everyonehas the luxury of a full-blown testinglab or the available users for applica-tion performance testing. Some soft-
ware manufacturers produce load-testing software from applicationvendors themselves to third-partyproviders.
PERFORMANCE TESTINGCONSIDERATIONSThe following are some important fac-tors to consider as well as questionsto ask before you begin virtual appli-cation performance testing:
� Host affinity.What are the applica-tion vendor requirements, if any? Canyou run one application on the samehost—or even in the same DistributedResource Scheduler or high-availabili-ty cluster—as another application?Host affinity is a requirement in someinfrastructures, and in others it justmakes good sense.
Application perform-ance testing is anessential part of anysolid virtual applicationplan, particularly fortier-one andmission-critical applications.
� Resource affinity.Again, is this avendor requirement? It may be neces-sary to dedicate quantities of the corefive resources—CPU, RAM, disk, net-work interface cards and USB—toensure the best performance and sup-port for virtual applications.
� Shares and resource pools. Theseresources don’t enter into the equa-tion until the end of the process,but you should consider resourcequantities up front. Determine theresources your guest machine andits virtual applications will use. Don’tnecessarily leave the hypervisor’sresource allocation at the defaultamount. You can fine-tune sharesthroughout the application perform-ance testing process to see what’sbest for each virtualized application.
� Dedicated clusters. To improveperformance and management, tryforming a cluster with an isolated setof servers that host common or inter-acting applications.
� Application stacking.With theultrafast and capacity-heavy hoststhat exist today, it’s possible to stacktwo or three applications on one virtu-al server. But with bigger applications,I advise against it. Stacking can be agreat thing, but it can also be a prob-lem if virtual applications don’t coop-erate with one another.
� Inter-application cooperation. Someapplications just don’t play well withothers in the sandbox. We know thatall too well from the physical serverworld. Some virtualized applicationswon’t interact well with others on thesame guest, host or cluster. And somejust cannot be virtualized at all.
As you begin virtual applicationperformance testing, create a baselineof servers and a load that’s as similaras possible to what’s on the physicalserver. It’s not always possible to testvirtual applications with 100% accu-racy, but you should be able to getclose.Also, keep in mind that some prob-
lems might emerge only after testingis done and you compare the data. Ifyou think of application performancetesting as an art, you’ll find some cre-ative ways to boost performance. �
15 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
Somevirtualizedapplicationswon’tinteractwell with oth-ers on the sameguest,host or cluster. Andsome just cannot bevirtualized at all.
UNLOCKTHE VALUE OF THE CLOUD�Drive innovation. reDuce costs. improve Business agility.
Our vision is to build not only a better cloud, but a more open one. We know your infrastructure includes diverse assets from multiple vendors. Our approach means you can manage these assets as one cloud. Without lock in. We make the move to the cloud an evolution, not a revolution.
www.dlt.com/unlockingcloudWc
17 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
ANY IT depart-ments are feelingan extraordinaryamount of pres-sure lately to
move applications to the cloud. Thispressure often comes from those whoare anxious to capitalize on the vastsavings advertised by cloud serviceproviders.Although there is no denying that
running applications in the cloud cansave money—especially in the formof startup costs—some organizationsare beginning to discover that thecloud experience isn’t always all it’scracked up to be.As cloud technology begins to
mature, security and availabilitybecome higher priorities. Andalthough performance can be just asimportant, it is sometimes overlooked.The bottom line is that your users
neither know nor care where an appli-cation is running. They expect theapplication to perform just as well in
the cloud as it did when it was hostedin your data center.
MONITORINGPERFORMANCE IN THE CLOUDSo how can you make sure that yourapplications are performing adequate-ly in a cloud environment? This simplequestion can actually be very difficultto answer. Odds are that the perform-ance monitoring tools that you usedwhen the application was being host-ed in your local data center will notwork in a cloud environment.First of all, you can’t depend on your
cloud service provider when it comesto performance. Typically, cloud serv-ice providers will not even addressthe issue of performance within theirservice-level agreements (SLAs).These types of agreements usuallycover service availability but notapplication performance.There are some cloud providers
that will give you performance bench-
MeasuringandOptimizing theCloudEVALUATE PERFORMANCEDATA IN A CONSISTENTMANNERWITHTOOLS SPECIALLYDESIGNED FORTHE CLOUD. BYBRIENM. POSEY
M
18 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
marks for your hosted applications.The problem is that this benchmark-ing data probably does not accuratelyreflect the end-user experience.Think about it for a moment: Have
you ever heard a cloud service
provider advertise that your applica-tion is going to perform better than itdoes now once you move that appli-cation to the cloud? Of course not.Cloud service providers know that nomatter how good their data centersare, there is nothing that they can doabout Internet latency.So if a cloud service provider gives
you performance benchmarking data,it’s likely that the data was collectedfrom within the service provider’s owndata center. If that’s the case, it proba-bly does not take Internet latency intoaccount.Even if a service provider does try
to account for Internet latency, geo-graphic proximity to the data centercan make a huge difference in end-user experience. Unless a service
provider actually measures your appli-cation’s performance from your facili-ty, then the performance data that itprovides probably won’t accuratelyrepresent true performance.So how can you monitor the per-
formance of your applications that arerunning in the cloud? The truth is thatthere is no one-size-fits-all solution.Each cloud provider exposes differentparts of its network infrastructure. Asa result, a solution that works withone cloud provider may not work withanother.
THREE TYPES OF CLOUDSYour options for monitoring cloudperformance depend on what type ofcloud you are using. The three maintypes of clouds used for hosting appli-cations include:
� Infrastructure as a Service (Iaas)—When an organization subscribes toan IaaS cloud, it is typically givenaccess to a virtual server and a stor-age pool. IaaS clouds such as Ama-zon’s EC2 provide the greatest oppor-tunities for performance monitoringbecause you have full access to theserver’s operating system and to theapplications that are running on it.
� Platform as a Service (PaaS)—Clouds such as Google App Engineare platforms for runningWeb-basedapplications. Although it is sometimespossible to build performance metricsinto aWeb application, PaaS cloudsdo not provide access to the under-lying operating system.
Cloud service providersknow that nomatterhowgood their datacenters are, there isnothing that they cando about Internetlatency.
19 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
� Software as a Service (SaaS)—These types of clouds allow organiza-tions to run commercial SaaS applica-tions such as Salesforce.com or Mi-crosoft Exchange in a hosted environ-ment. Although the hosting providermay give you access to performancereports in some cases, there are fewoptions for gathering your own per-formance metrics.
If the application that you need tomonitor was developed in-house, thenone of the most effective techniquesfor monitoring the application’s per-formance is to have its developerscreate an agent that can be embed-ded into the application. The general
idea is for the agent to note the timejust before phoning home. When theagent’s transmission is received, thetime stamp on the transmission canbe compared against the time whenthe transmission was received, whichwill allow you to measure latency.
TOOLS YOU CAN USEIf you happen to have an applicationthat is coded in Java, you don’t haveto worry about the hassles of develop-ing a custom agent. There are at leasttwo software publishers that offerJava-based code for monitoring cloudperformance. One is AppDynamics,which is designed to monitor
MONITORING THE performance of applications in the cloud is important to ensurethat end users’ needs are being met. But there are other reasons why you shouldmeasure cloud performance.
One of the main objectives that draws organizations toward hosting applica-tions in the cloud is scalability. Running an application in the cloud frees an organi-zation from dealing with the constraints of physical hardware. Some cloudproviders are able to dynamically allocate additional hardware resources to appli-cations to help them perform better during peak usage periods.
But as nice as dynamic scalability is, cloud providers tend to charge customersbased on the hardware resources that they use. This means if you allocate exces-sive resources to your hosted application, you will end up paying more than youneed to. Alternately, if you don’t allocate enough resources to the application, itwill perform poorly.
By monitoring an application’s performance and comparing the performancemetrics to the application’s usage statistics, you can determine at what point addi-tional resources need to be dynamically allocated. Finding the optimal hardwareusage levels will help to save your organization money while ensuring that applica-tions perform as expected. �
TAKING ITTOTHENEXTLEVEL
20 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
distributed Java-based applications inthe cloud. The software tool providesa dashboard for tracking applicationperformance and helps to isolate thecause of any problems that may bedetected. AppDynamics also detectsmemory leaks in real time and locatestheir sources.Another tool that exists for moni-
toring applications in the cloud is NewRelic, which works with Ruby, Java,PHP and .NET applications. Two fea-tures make New Relic unique. First, itis designed to be simple to deploy.The other is that the basic monitoringsolution is free. If you also need diag-nostic and optimization capabilities,New Relic offers those tools for a fee.AppDynamics and New Relic are
suitable for use with IaaS and PaaSclouds, but they do not work withSaaS. But another application calledFactFinder from BlueStripe Softwareis said to work with any application,regardless of whether it is a packagedapplication like Microsoft Exchange ora custom app that was developed in-house.FactFinder works by using protocol-
level visibility into each node that isinvolved in running a distributedapplication. By monitoring packets ona hop-by-hop basis, FactFinder is able
to create a map of the infrastructurethat is running the distributed applica-tion and locate bottlenecks when per-formance problems occur.
Regardless of which method youuse to collect performance data foryour cloud-based application, it iscritically important that you measurethe performance data in a uniformmanner. Unless you are consistentwith your choice of metrics and thetools that you use, the historical per-formance data you collect will bemeaningless. �
Regardless ofwhichmethod you use tocollect performancedata for your cloud-based application, itis critically importantthat youmeasure theperformancedata ina uniformmanner.
Building Trust Around The Globe
When you want to establish trusted relationships
with anyone, anywhere on the internet, turn to Thawte.
Securing Web sites around the globe with:
• strong SSL encryption
• expansive browser support
• multi-lingual customer support
• recognized trust seal in 18 languages
Offering outstanding value, Thawte is for those
who know technology. Secure your site today
with a Thawte SSL Certificate.
www.thawte.com
© 2010 Thawte, Inc. All rights reserved. Thawte, the Thawte logo, and other trademarks, service marks, and designs are registered
or unregistered trademarks of Thawte, Inc. and its subsidiaries and affiliates in the United States and in foreign countries. All other
trademarks are property of their respective owners.
22 VIRTUAL DATA CENTER E-ZINE APRIL 2011
EDITOR’SLETTER
7
PUBLICVS.PRIVATE:
COMPARINGCLOUD
ARCHITECTURES7
HOWTOSETPRIORITIES FORVIRTUALIZINGAPPLICATIONS
7
MEASURINGAND
OPTIMIZINGTHE CLOUD
7
Stephen J. Bigelow, a seniortechnology editor in the DataCenter and VirtualizationMedia Group at TechTargetInc., has more than 15 yearsof technical writing experi-
ence in the PC/technology industry. Heholds a bachelor of science in electricalengineering, along with CompTIA A+,Network+, Security+ and Server+ certifica-tions, and has written hundreds of articlesand more than 15 feature books on com-puter troubleshooting, including Bigelow’sPC Hardware Desk Reference and Bigelow’sPC Hardware Annoyances. Contact him [email protected].
Mike Nelson has been in ITfor more than 20 years, withexposure to a diverse field oftechnologies and solutions.He has devoted more thanhalf a decade to virtualization
and server-based computing. Currently,Nelson is a senior analyst at a Fortune 100company in the midwest United States.
BrienM. Posey has receivedMicrosoft’s Most ValuableProfessional award seventimes for his work withWin-dows Server, IIS, file systems/storage and Exchange Server.
He has served as CIO for a nationwidechain of hospitals and healthcare facilitiesand was once a network administrator forFort Knox.
ABOUT THE AUTHORS
Cathleen GagneEditorial Director
Christine CasatelliEditor
Linda KouryDirector of Online [email protected]
Marc LaplantePublisher
Peter LarkinAssociate Publisher
TechTarget275 Grove Street
Newton, MA 02466www.techtarget.com
© 2011 TechTarget Inc. No part of this publicationmay be transmitted or reproduced in any form orby any means without written permission fromthe publisher. For permissions or reprint informa-tion, please contact Renee Cormier, Director ofProduct Management, Data Center Media,TechTarget ([email protected]).
SPONSOR RESOURCES
See ad page 3
• Best Practices for Virtual Infrastructure Management
• Data Center Evolution Without Revolution
• Simplifying the Virtual Data Center
See ad page 11
• Move fearlessly among the clouds. Evolve your data center. Transform your business.Discover better ways to work. Together.
See ad page 16
• Cloud Infrastructure Matters
• Red Hat Cloud Foundations: Cloud 101
• Unlocking the Value of Cloud On-Demand Webcast
SPONSOR RESOURCES
See ad page 21
• thawte products
• free thawte guides
• thawte live chat