Private Cloud Is Not A “One Size Fits All” Solution
I recently had a discussion on the use of private cloud, specifically on vCloud Director, over traditional server virtualization on vSphere. Essentially – should all workloads fit into a cloud model, or do some still merit the use of a traditional server virtualization. I’d argue that “standard” virtualization is here to stay, and that private cloud is not a one size fits all solution for all workloads. In some situations, there isn’t really any gain in going to a cloud platform, as it’s more than just a technical solution and requires a lot of process overhauls from the various teams that utilize the cloud (including the business itself).
Peeling The Onion
Unless you are realizing a specific gain from having a virtualized data center inside of your already virtualized vSphere data center, an additional cost has been added to get the same results. 
vSphere is already great at handling multi tenancy in the form of network and storage IO controls, resource pools and reservations, limits, shares, and clusters. The point of private cloud is to expose a subset of your environment to the internal client or department in a self-serve manner that also prevents them from hanging themselves (especially on the network part). If you’re not doing that, just stick with vSphere.
This isn’t to say that a private cloud (vCloud Director) doesn’t bring value to the table. I’ve found that vCloud has some really cutting edge networking features available to it. And really, the network is both the most complex and powerful part of the product. It takes genius to design a system that can use the same set of IPs all over the place while using VLANs or MAC-in-MAC encapsulation to keep things working.
Question is – are you going to use these features?
The Wild West
This may hold especially true for those worried about departments pushing out production workloads, as I tend to believe those are best served with some level of systems administration (read: IT department) interaction. From my experience, developers rarely make good server admins for a variety of reasons. This isn’t a dig on developers – these are some smart folks – but I tend to see a “get it working and move on” approach, and a general lack of experience on the proper way to build and maintain a system (more so in the Windows world).
The private cloud can definitely handle production workloads. This isn’t in question. But, should it? The private cloud is all about agility and flexibility, in the form of being able to hand out resources quickly and with little administrator input (after the original constraints are set for the organization). Having production workloads that will exist for a long period of time stagnate in a private cloud and tie up resources that could be better serving other dynamic, volatile projects. Shorter lived workloads make more sense to me.
Thoughts
I think private clouds are very interesting, and a trend is building towards getting everything inside to live happily. It’s always worth taking a look around to ensure that the use case is met, and the functional requirements are satisfied. This is an open discussion, and agree or not, I look forward to your comments.
Very interesting comments and I do agree. There needs to be a lot of thought put into why a business wants to implement a private cloud. Some IT departments may feel that some of the work of deploying servers will be passed onto the users but I think in the long run most devs (not all) will end up causing more of a mess which IT needs to fix. Personally I’d rather be involved in helping to set up an environment than having to come in and fix it. The flip side of my own argument is that whatever they mess up will be isolated to their own vDC
Thanks for the comment, Mike. I view a properly tuned vApp, with limited and controlled user power to alter, as something that should alleviate a lot of developer-cowboy issues. A good vApp build requires a lot of server admin / engineer work up front, with a user simply control the “quantity button” on the front end.
In practice, it has been much more complex to build a robust and modular vApp that can be rapidly provisioned than it would be to spin up an environment here and there.
I couldn’t possibly agree more with you. It sometimes seems to me that people only deploy vCloud Director because “cloud” is the recent koolaid buzz everywhere, without having actually analyzed their requirements first.
I personally see vCloud Director as a tool primarily suited for the business model similar to an internet hosting provider (like dreamhost etc) with fully self-sufficient tenants, who have no ties or respeonsibilities whatsoever to a central “IT department” above.
All the cool networking stuff you mentioned brings little to no value in most companies internal networking infrastructures. Another issue is that sometimes they don’t (want to) implement a real chargeback process for internal uses. Brace for VMSprawl at it’s finest.
Hah! I like the sprawl part of the comment. Although, with vCloud it’s much easier to control sprawl with lease timers and such, so the sword can cut both ways. If left wide open (unlimited leases) I can see resources being consumed quite rapidly with no recourse.
Private cloud is anything in your own data center. For example I have installed storegrid professional edition in own data center and so it acts as a private cloud. I am replicating the data to amazon s3 and so it acts as Hybrid cloud setup.
I agree with you, Chris. I tend to do a lot of VMware & vCloud Director consulting in my day job and I’ve found that dev/test is usually the first thing that folks want to address with private cloud. It’s an easily understood use case, and serves as a great way to introduce parts of the business to the concepts around self-service, chargeback, etc. Additionally, lessons learned during the course of service design & implementation may help provide valuable inputs to additional service design methodologies.
It never really made much sense to me why folks would want to run static production workloads in a private cloud. There are almost no advantages in terms of availability and may (in some cases) only serve to complicate the lives of traditional IT folks.
Chris, very interesting post and something I’ve been trying to get across to people. I also did a post called “Why VMware vCloud Adire for isn’t for odays private cloud”
http://www.wooditwork.com/2011/06/03/why-vmware-vcloud-director-isnt-for-todays-private-cloud/
I find people are often buying private cloud products for the wrong reason, often just for the automation part and not really for the rest of the service orientated approach.
Way is interesting though is I believe VMware wants t address this with its DynamicOps acquisition. This is a fantastic automation product without the network capabilities or complexity that vCD has which is exactly what many people actually need.
Whether static production VMs are suited to a private cloud is an interesting discussion but in my opinion you want everything in one place if you’re going to be doing any chargeback.