Lori MacVitte over at DevCentral posted an interesting article on virtual server sprawl. I’ve seen this topic come up over and over again on panels, in the press, on blogs, and in customer conversations. What’s virtual server sprawl? It’s the theory (or fact) that because virtual machines are so easy to create that once you virtualize your servers will start to multiply like rabbits. I’m here to tell you there’s a lot of fact behind that statement. After doing virtualization for 6 1/2 years here at VMware I don’t think I’ve seen one customer reduce their server count ever. Sure, they’ve gotten rid of a lot of physical hardware. I guess you could call those servers. However, I’ve never seen the actual server count (OS/App stack) go down. Usually it starts to ramp up rather quickly. There are various reasons for this.
In no particular order the reasons I see for virtual server sprawl are the following:
1) Customer have no clue how many servers they originally had. I see this one over and over again. Usually every virtualization project starts off with a capacity analysis of some sort. Basically you scan the network, find servers, and then query the servers for basic information like OS, Apps, and utilization. There are lots of tools and companies that do this including VMware’s own Capacity Planner. Before I start a capacity analysis I usually ask the customer how many physical servers they have (it’s also a good way to qualify an account). What amazes me is I get different answers for the same environment depending on who I ask. That’s not too surprising. In a physical environment you often have rogue servers out there. You have servers under people’s desks, servers in the broom closet, and servers in the bathrooms (yes, I’ve seen that too). You really have physical server sprawl. What this means is you think you are starting with 300 physical servers but after you consolidate you have 480 servers. It’s the rogue servers that put you over the limit and make it look like you have virtual server sprawl.
2) Virtualization frees up physical resources. Some customers are virtualizing because they’re out of power or out of datacenter space. This lack of resources is limiting the growth of the company and possibly holding up very important projects. Once you go virtual you now have extra capacity in the same footprint and so servers start getting created like mad for the backlog of projects that are waiting. This makes it look like you have server sprawl but really you’re just implementing what you’ve been waiting to implement and so it’s actually valuable for the organization.
3) Virtualization frees up human resources. No, not HR – technical admins. A good but of an engineer’s time is sometimes spent ordering hardware, tracking it down, getting it racked, getting it cabled, getting it inventoried, etc. I heard a lot of groans on that one.
These are very low value tasks for the highly capable and often highly paid admin. By virtualizing you reduce these tasks to a bare minimum. The new free time can now be spent on higher value tasks such as helping business units get their apps and products to market faster. This results in some server sprawl since again new projects are getting done that were being held up for lack of resources before. Of course that new time also ends up being spent with family and on the golf course as well.
4) Virtualization allows for redundancy. As one of the commentors on Lori’s blog mentioned, virtualization allows for some redundancy where you didn’t have it before. Now that you’ve virtualized 10 or 20 servers to 1 physical platform a lot of customers choose to implement some redundancy for their virtual servers. Sometimes this is as easy as enabling VMware’s HA feature. Sometimes this means setting up clustering to another virtual or physical server in the environment. The latter means more servers in the environment that weren’t there before.
Whatever the reason for server sprawl it isn’t necessarily always bad. It’s not like these servers are getting created automatically by some ghost in the machine. Most of the time trained admins with the correct permissions are creating these servers for very real reasons. As long as you have a good, trusted, trained admin at the helm virtual server sprawl is not something to worry about.
Tags: VMware


October 2nd, 2008 at 10:52 am
Another reason for apparent virtual sprawl:
When using physical servers, sometimes a small app will get installed on an existing server alongside another app, to avoid buying new hardware. When you virtualise such a landscape, those apps typically get split onto separate virtual machines. And in future they’ll get their own VM from day one.
October 2nd, 2008 at 3:49 pm
Mike, great summary on virtual server sprawl! Tools like VMware Capacity Planner and our software, Lanamark Suite, enable our partners to assist customers in reducing physical server sprawl. And products like Embotics V-Commander ( http://www.embotics.com ) can help reduce server sprawl once these servers are virtualized.
October 8th, 2008 at 7:17 am
Mike, great article indeed. I fully agree with your points. Perhaps something to add to the definition of vServer Sprawl. To me server sprawl has a negative connotation in that it implies uncontrolled growth.
The points you mention indeed lead to increased virtual servers, but this increase is due to legitimate business reasons. Certainly the pent-up demand of servers is a valid one, whether separation of workloads into their own OS is legitimate in all cases (as Steve commented) is probably up for discussion.
In any case, the problem(?) exists and to me it is indeed a problem if/when there is a lack of control. However, where there are problems, there are also solutions (certainly various ones available to address this sprawl problem).
It is interesting to me that we speak of virtual server sprawl as it is a sign of where we are in terms of maturity on the adoption curve of virtualization.
October 14th, 2008 at 4:55 am
It is interesting to me that we speak of virtual server sprawl as it is a sign of where we are in terms of maturity on the adoption curve of virtualization in Virtual dedicated server.