Hardware

Should every server be virtualized?

Today's IT environment is a mix of technologies that are clearly moving faster than the typical IT environment can handle. Virtualization expert Rick Vanover shares a few points for and against approaching virtualization broadly.

Frequently, I meet with people who are considering virtualizing servers in a number of different scenarios. These can be organizations that are just getting started with virtualization, which can be a big change in light of shrinking budgets and limited supporting infrastructure such as shared storage. Another situation is virtualizing the difficult servers. The low-hanging fruit was easy for early adopters to virtualize the datacenter, but there are plenty of more challenging systems left. This includes servers with a large amount of local storage, limited amounts of downtime or systems that are incredibly sensitive. The last situation is the organization that is on the brink of being 100% virtualized.

In all situations, the question comes up if everything should be virtualized. The short answer is of course, “It depends.” However, in reality only in a few situations is this really attainable. The enterprise, of course will have a number of systems that should not be virtualized for any number of reasons. This can include vendor support requirements or an alternative arrangement such as a clustering solution to provide additional reliability.

While virtual machines are all fine and dandy for just about every workload, physical servers are still “okay.” I get stuck coming up with virtualization solutions for very small environments. In those situations, I would gravitate to a free virtualization solution such as the free edition of VMware ESXi (vSphere Hypervisor) or Microsoft Hyper-V. But even then, it may be easier to simply go with a physical server for a site that only needs one or two servers.

It is definitely nice to tote around a statistic such as, “I am 100% virtualized.” But, most organizations need some sort of additional qualifier. In my practice, I usually quantify server virtualization inventory in terms of “this datacenter is 90% virtualized for all eligible workloads.” This means that systems that are not supported as a virtual machine by the application vendor or other justified reason are not counted in the calculations.

The issue that I run into most is that our gatekeeping procedures seem to be slipping. I’m totally fine with the process changing, which is the case with the ever-increasing capabilities of virtual machines. But, there are still today solid reasons on why servers should be installed natively on hardware, and we should ensure that this process is still followed.

Do you run into pressure to make everything virtual? What guidelines or boundaries do you put in place regarding virtual machine candidacy? Share your comments below.

About

Rick Vanover is a software strategy specialist for Veeam Software, based in Columbus, Ohio. Rick has years of IT experience and focuses on virtualization, Windows-based server administration, and system hardware.

41 comments
bbnetman
bbnetman

Making recommendations and decisions for the future: Recommendations and decisions for the bank???s best use of money and resources that I have made are based on some of the following criteria: ??? Does the technology interface with our core application provider and does it make sense with the current environment? ??? Does the technology eliminate or add a single point of failure that would affect users on a global scale? ??? Does the technology strengthen or weaken our overall protection and disaster recovery plan? ??? What is the risk and business impact of the technology? ??? Does it make financial and technical sense now and in the long term when taking all factors into account? Factors would include security, network integration, licensing, training and hardware costs and maintenance. ??? Has all due diligence and research been completed to implement the technology? ??? Does the technology align with the bank???s business strategy and what are the future expectations of the technology? In our environment virtualization would add the benefit of some consolidation but I do not believe the advantage would be a good resource of the bank???s money and time at this moment. The reason for this is: ??? If we were able to virtualize all of our servers into a single server we would create a single point of failure for the bank. That failure, or disaster, would be in terms of a data circuit or server malfunction that would lead to a disruption of all services to bank employees. If the data circuit failed where the physical virtual server existed the bank???s users outside that office would not be able to access anything in the form of email, word/excel documents, shared folders, network printers, and any other network resource that may be available on that server. If the virtualized server itself experienced a critical disaster it would completely disrupt all access to the server. A critical disaster could be a hardware failure, a security event or damaged files on the server. ??? Our core application system, along with pther main core products are already serviced and accessible in a dedicated cloud. Virtualization would not add any benefit to these main products of the bank. Even if we still had these core products in house, virtualization was not supported by our core vendor when I checked then and still is not to this date. ??? We have some application servers that are not supported using virtualization. These servers would still exist in the environment greatly minimizing the advantage of consolidation. ??? New servers, hardware and licensing would have to be purchased once a virtualization platform was decided on. There are a few different virtualization platforms to choose from and all would cost the bank a substantial amount of money and time up front to implement. ??? Having our core application and other main applications in the dedicated cloud combined with other applications on servers being incompatible with virtualization would lead to a small footprint to actually virtualize. ??? We could virtualize our few file servers and Active Directory servers but it would not be worth the costs and would create a single point of failure in the event of a disaster. ??? Virtual servers are a huge benefit when there is only one physical location of business. Having multiple locations raises the risk of down time due to data circuit failures. ??? We have two sites that currently have the bulk of the servers. If we went with a virtual server system at each location it would have little advantage to the other three sites that have only one server to begin with that could be moved to a virtualized environment. ??? We are already on the path of server consolidation. By moving our Core Application to the service bureau, along with our main applications, we have eliminated around 6 to 8 servers combined with the addition of backup servers at an alternate branch location. I expect that number to grow with expanded dedicated cloud offerings by the core vendor. We are also in the process of finally shrinking our legacy image platform by decommissioning them.

adminmichael
adminmichael

I believe that if you creat a virtural server it has its ups and its downs yes i believe it can be very spendy but i also believe that they can be very helpful as far a getting a new one up and running very quickly!

Raulf123
Raulf123

Virtualization vendors are very good at pointing out why virtualization is good for you, and they are usually right. Let's try to compile a list of exceptions, starting with the 3 mentioned by Rick: * Servers with LUNs over 2TB (current ESXi limitation) * Servers that need continuous availability * Systems very sensitive to Network/SAN delays * Servers running constant-load applications at 90%+ CPU (e.g. video recorders) * Servers that are load balanced (N+1) to reach full utilization * Servers running applications that need > 50% of CPU for each instance Any others? Raul Fliman Senior IT Architect

herauthon
herauthon

Sometimes an external NAS coupled with a decent SQL system can be the thing if one looks at replacement - variables: restore time, restore cost, implementation, availability of hardware, learning curve of employee, basic warrenty of goods. It's more easy to buy redundancy if one can obtain it from the next corner ???

jhoward
jhoward

Some things virtualize very well such as web servers and other servers that do distributed tasks such as media streaming. The ability to add memory, CPU and rapidly clone these servers for clustering, load balancing and DR is invaluable. Other things such as highly transactional databases and VoIP PBXs do not transition to virtualization so well although this is becoming less the case as virtualization becomes more of a development target. The fact is that today you would be pretty hard pressed to find an organization (or at least users within that organization) that doesn't use virtualization in some way - either test environments or even running Windows/Mac/Linux on the same PC at the same time. Overall as the hardware required for virtualization becomes more of a commodity, virtualization will become even more popular than it already is.

chabney
chabney

I agree with the previous post that the article doesn't contain as much substance as I was hoping for when I saw the title. What disturbs me more, however, is that it seems to me that most of the people who have written the "negative" remarks toward virtualization above, may not have good virtualization experience to begin with. (either that or they just don't understand the technology they're working with!?) Sad, because since the article itself doesn't have as much meat as it should, quite a few people will end up gaining as much / more information from these posts as they do from the article! :-(

laseats
laseats

You do realize that article said absolutely nothing in 434 words -- right?

Alpha_Dog
Alpha_Dog

That's like saying all of our company vehicles are economical subcompacts. Sure, it's nice to say, but what happens when you need to move a truckload? The point isn't capacity, but rather that each situation is different even within the same organization, and not all situations merit virtualization. Got 8 servers with an average of 20% utilization and need a 9th? Virtualize what you can and save a bundle. Need a firewall/VPN box, fileserver, and app server? You will want the firewall VPN appliance to be physically separate from the rest of the infrastucture. Virtualize to put two highly accessed applications on one machine? Bad idea. Use 2 machines and virtualize the failover machines for the fileserver on the application server and vice versa. with replication set as a "nice" process. Virtualization is but one tool in the IT toy box. Use it where the application fits, not for bragging rights at golf.

reisen55
reisen55

A co-consultant I work with, a certified BCP-DR planner, discussed virtual servers and the big flaw in the argument is that what if the PHYSICAL BOX itself goes? Hardware failures do happen and if you drop six servers in one server because of a failure, THEN you have lost the lot. I am in favor of virtual servers for (a) redundancy and (b) secondary usage systems but always keep your PRIMARY SERVERS on individual systems so that points of failure are reduced as much as possible and restoration times decreased accordingly. WHY TAKE A CHANCE?

ilya.shick
ilya.shick

It is not silver bullet to get 100% or 75% VM's or.. It is not OK to say "physical servers are still okay". It is a misleading statement after all. It is depend of flexibility, performance/availability and technical support. As first priority be a "Hybrid" structure, architecture on Data Center.

patmura
patmura

As a VAR we've run the numbers for our clients to go to the Cloud. It doesn't pen out. The upfront costs are lower, but in the long run the net cost is always higher...much like leasing server hardware. If something goes awry, the end-user has absolutely no control or alternatives other than to sit and twiddle their thumbs until an unknown entity in another state addresses the problem. I will never advise my clients to relinquish control of their confidential and sensitive data to virtualization.

blarman
blarman

I liked best the point about how small outfits with only a few servers really don't benefit much by virtualizing. That's my company in a nutshell. We only run two database/application servers and a file server. The cost to virtualize all of this would have doubled-the cost of our datacenter without really bringing benefits even close to the costs. It's not that we couldn't, only that there is no payoff in it for us. Between the licensing costs of the virtualization software (and the training to use it), the additional hardware to put it on, not to mention the additional licenses for server software, etc....

Jack Flash
Jack Flash

Hey Rick, It is all a matter of what is the goal. Once the goal is efficiency and savings of time / money it all falls in the right slot. This results in deploying Virtualization when you need it, rather than just as a promotional goal. On the other hand sometimes managers will deploy Virtualization for non-practical reasons. Sometimes it even has a positive result, if it paves the way for people who were sitting on the fence, provided practical decisions are taken at some point. This directly hooks to the recent article I posted...here http://itprofessional-mastermind.com/blog/server-consolidation/server-consolidation-the-cloud-people/ Yours, Jack.

dbarlamas
dbarlamas

We are a small health care orgainization that is 75% virtualized. One application in particular, our main clinical, financial and HR package, is seeing performance hits that we are trying to deal with. The VMware is great for DR and the flexiblility. Fine tuning the performance of the application, fine tuning the OS, then fine tuning the VM can be a dauting job. I am struggling with this. Particularly with the understanding of how the application utilizes the cpu for the instructions as to how to improve the processes. It is easy to put these machines together and even with the statictics of Vcenter and knowning that the VM is not struggling and knowing that you don't have network issues, finding the right balance when you know there is performance problems, is a hard job.

herauthon
herauthon

One could also ask, why not undo virtualization if it doesn't improve service, if it increases overhead and maintaince cost etc. Or why would a company go for virtualization if they can export the management to dedicated systems by other hosting companies? Is datastorage part of the internal structure of every company?

b4real
b4real

The organization needs it now - so virtualization is the best way to get it done!

b4real
b4real

Will cause virtualization to be unfeasible. There are a few titles that take all compute resources of a cluster into consideration for licensing. For example, a database application may require all CPUs to be licensed - even though no more than 4 can be used.

gcomputeronet
gcomputeronet

If you have heavy networking traffic for a server, it won't do better being virtualized as it now shares the nic with other servers.

mattlieblong
mattlieblong

As I commented below, very small environments with only one or two servers. I want the benefits of say ghetto backup scripting, or other whole-vm backup/restore ability, but need it all to run in and out of one server. Want a complete physical hardware failure to result in some quick method of restoring an entire VM from backup to a new machine and pushing play.

Jaqui
Jaqui

[b]"Servers that need continuous availability"[/b] with most virtualization options they have a high availability feature that removes that point from your list. The availability issue was brought up long ago and the options for virtualization technology added support for that to become more usable.

b4real
b4real

Consider the task of going to the corner to get the parts. I'm a believer in abstraction within a compute cluster. If you can do that, then your RTO is shrunk.

b4real
b4real

Hi chabney: I wrote the blog post. I intentionally left it vague to get people talking about what virtualization means to them. I'm "ALL IN" virtualization myself, but yes, sometimes physical servers make more sense. As for lacking experience with virtualization - yes, there are plenty of people whom are starting at zero with it. I am working to help people embrace virtualization (Side note - see my virtualization blog posts in the servers and storage blog on TechRepublic or rickatron.us). The best way to embrace virtualization is to first have it shaken down by real users like you and me.

b4real
b4real

The point of this post is to get people talking.

watsonderekj
watsonderekj

In a truly virtualized environment the virtual servers are stored on a SAN unit and they are loaded into a cluster of computers to perform the work. From what it sounds like he just has one server that is hosting several guest operating systems internally. Yes this is virtualization but it is not real virtualization, more or less just consolidating the equipment.

b4real
b4real

HA, FT, clustering... I can do mission critical virtualization any day of the week.

gcomputeronet
gcomputeronet

There are servers that should not be virtualized, or should be evaluated carefully. I've seen way too many virtualized solutions that have too many servers for too little hardware. With each vm server fighting for physical hardware use everything slows down. Here where I am working they lost the test physical server, so all test virtual servers had to be rebuilt. That took two weeks.

GSG
GSG

The point is that you shouldn't have one point of failure. We have a solution where if one bank of VM's go down, then we can transfer the VM to another bank. In some cases, this would be an automatic process. Think of it as clustered VMs. We had some VM's but we've replaced those and have the Vblock. I think the key to this is to have someone who really understands the concept and is very meticulous about which VM solution they use and how they manage it.

mmgrady
mmgrady

You need to consider that virtualization provides redundency. A proper configuration where that is the capacity to restore all of the required processes from one (sometimes more) failed servers is critical. Even when the remaining servers could have performance issues (slow response), that is much better than no service. The spliting of the hardware into different seperate pieces (Server or blades and SAN's) provides the ability to grow. I have found that a SAN and bladecenter provides a great deal of recoverability well beyond what any stand alone system can provide. In nearly every case you can have hot swap capability and maintain the systems up time. Even datacenters of only 2 or more servers can benefit. They normally have limited IT support, and virtualization can be configured to automatically recover servers and processes without the IT support being there.

ederkley
ederkley

It always comes down to the value of your data/uptime versus the cost of the solution. The particular issue the co-consulant mentions seems to be easily mitigated by having at least two physical hosts for providing runtime resources of CPU and RAM and a separate SAN for data storage, at minimum. As an example, VMWare in that environment would allow you to quickly restart the virtual servers that were previously running on the failed host on to working host within minutes. Sure if all your servers suddenly have to work on a single host then performance may be degraded across the board but critical resources are still available very quickly, leaving you free to fix the issue instead of trying to get critical resources back online some other way. This can be done without any additional licensing for vCenter, however vCenter is also useful for things like live migration, automated high availability, ease of management etc. Again, this all assumes your data is actually kept on a SAN and the physical hosts are just for providing runtime resources. Server-grade hardware and SANs are highly stable devices and feature multiple redundancy options for power, controllers, etc to help reduce the risk further. And then you could also have redundant SANs altogether if your data and uptime requirements are so critical...

brent.young
brent.young

If you have a bunch of ESX servers hosting all your servers and one goes down, those VMs are moved to an operational server faster than one can recover the same server on a physical box.

b4real
b4real

It is absolutely OK to say physical is still OK today. But as in every situation, it depends on the situation and its requirements.

VBJackson
VBJackson

At least not as used in this article. Granted, most Hosting centers and "The Cloud" heavily use virtualization technology, but in this article the author is talking about hosting multiple OS instances on one physical server, which may (is probably) located on-premise. Not that your comment isn't true, it is just on the wrong topic.

Jaqui
Jaqui

in the article where he mentioned the FREE [ zero licensing cost ] options for virtualization technologies. that makes you cost assesment off. ;)

b4real
b4real

You can provision a lot of network resources to a VM. Further, if the VM, application server and database server are all virtual machines and on the same host - they will use CPU and not traverse the physical media.

b4real
b4real

The obvious issue becomes the shared storage configuration as a single point of failure. To address that, there are plenty of advanced data protection solutions as well. (Cheap plug for Veeam!)

b4real
b4real

That the definition of the infrastructure is important. We have to know what we have to know what to do with it, virtual or not.

VBJackson
VBJackson

Your comment is true, but only IF you have "a bunch" of ESX servers, AND you don't already have them running at 90% capacity. Conversly, I am quite fond of running a pair of Hyper-V servers in a failover cluster with an iSCSI SAN. This setup minimizes the risk of down-time, and provides good performance for any applications that are not so heavily disk-bound that iSCSI doesn't provide the performance required. Would I put my main firewall on it? NO! Despite the expected segregation between VMs, I wouldn't care to trust anything except separtate physical hardware for something as critical as a firewall. And there are few things as disk-intensive as an LOB database, so I would want to take a VERY careful look at the usage statistics before I virtuallized it.

b4real
b4real

Being used in everyday discussions. Many enterprises are delivering awesome private clouds today, built on virtualization.

b4real
b4real

Here's a free puppy.

chabney
chabney

(1) The definition of "bunch" depends primarily on the size and number of VM's you need to auto-restore, and can be as few as 2!! This is at 50% utilization......maybe not "practical" at this number, but still only at (n-1)/n utilization for all other cases!!......which doesn't sound nearly as bad as simply saying "a bunch", imo!! (i.e. with only 4 ESX servers running at 75% utilitzation, I may be able to support a complete failure of one of those physical servers! considering that I may not have to worry about having any "test" and "dev" environments auto-restore, and I can probably even do better!!) (2) I find that the flexibility I gain with virtualization and the increased recovery time of my environment often offset the costs of any infrastructure improvements that I may need to perform in order to support it!! (3) Having seen virtual environments "certified" in regards to the segregation of data (even for evidentiary purposes!!), I don't find myself in the camp that says you have to buy separate physical equipment in order to effectively segregate data. You just need to know how to set up your environment!! :-)

lhAdmin
lhAdmin

So true! That's a great analogy.

Editor's Picks