Day to day, I work for a telephone company, though we are really much more – we sell cable TV, wireless, landline, Internet, security and home automation. [Disclaimer – these are my views and not theirs.] We have tossed around the idea of becoming a public cloud provider more than once. Last week, during HP Discover, I got my first full view of how we might realistically do that. We are not currently a HP CloudSystem customer, but I could see it fitting our needs.
HP unwrapped CloudSystem during the Discover conference in Las Vegas. CloudSystem is a full stack solution from storage, to networking, to servers and software to enable the cloud, whether it be private, public (service provider) or hybrid. So, where to begin on discussing CloudSystems – lets start with a definition.
What is the cloud?
To begin, this is a grey area. Cloud means many things to different people. Most agree, however, that cloud computing and virtualization are not the identically the same. In many people’s minds, the two are synonymous – mine included – until I really dug into the topic in recent weeks. I think the VMware, because it is pushing the cloud and because it is known for virtualization has helped further this incorrect concept.
The difference between cloud and virtualization is that cloud computing has a management software layer to enable provisioning of services. Cloud computing includes a level of automation and management. In HP’s solution, cloud doesn’t necessarily mean virtualization, although its heavily utilized. The key to cloud is automation.
Basis of the HP CloudSystem
From talking with an HP engineer in the HP Discovery Zone, I got a clearer picture of what makes CloudSystem tick. The core of the HP CloudSystem is the HP Server Automation (SA) software package and, optionally, HP’s Rapid Deployment Pack (based on Altiris). These two packages are designed to bring a level of automation to lifecycle of systems. SA is used to take requests from end users, create and build out the systems, add the software packages to the base OS, and deploy it to users. For lifecycle, SA is also used for patch management and change monitoring/mitigation. The HP engineer I spoke with showed me how an application stack is built out on a newly provisioned system and how applications are added after the fact from a service catalog.
The key here is that the server build process is no longer a manual task for administrators, but rather a standardized task performed by the automation software. The service catalog is key to making this happen and this is a mentality switch for many IT operations. Gone are the days of checklists and manual point and click build-outs. These are replaced with scripted installs for modular components.
The Service Provider’s Role
From a service providers perspective, there are two roles that we can play for customers. The first is that of a hosting provider – where we fully house their IT operation in our public cloud. The second is burst public hosting, or a hybrid approach.
The first model, the purely public cloud model, has been around as long as commercial web hosting has been available. Since shared hosting arrived in the 1990’s, this business model of a for-rent data center, rack or server has housed the majority of websites on the internet. The difference in cloud hosting, whether its Amazon EC3 or Terremark’s vCloud hosting, is the automation layer for customers.
For many customers, a public cloud only model may work. But for larger companies with established IT departments, there are many cases where data is too confidential to be hosted in a public cloud, either because of compliance rules or security concerns. This is where the second model, the hybrid or burstable cloud is coming into play.
The idea of burst is that companies can more “right-size” their infrastructures to handle their normal workload and then push additional resources into a public cloud for the times when their demand exceeds what they are capable of handling in-house. A good example of this might be a college or university. Computing resources required during registration increase drastically and as a result, the university spins up several additional registration web servers in the public cloud to accommodate the increased demand. After registration, the public extension is decommissioned and all computing brought back to their private cloud.
In some cases, such as steaming media providers, they may want a hybrid deployment with a source streaming server and redirector or proxy services in several public clouds to push the workloads out to be handled in several data centers and spread their overall workload. These are a couple really high-level (simple) concepts, but I think they help illustrate how this could work.
HP CloudSystem’s unique take
I had the pleasure of sitting down with Steve Dietch, VP of CloudSystem, at the Blogger’s Studio last Thursday as they discussed CloudSystem with a group of bloggers. HP has created CloudSystem as an open cloud stack. It is hypervisor agnostic, running VMware ESX, Microsoft’s Hyper-V or Citrix XenServer. And, you are no locked to a single choice per CloudSystem – you can mix and match hypervisors within a CloudSystem deployment, running two or all three side by side. To go even further, HP is supporting third party x86 servers and storage with their CloudSystem solution.
Unlike other vendor solutions, like vBlock, CloudSystem is not a single size solution. It is a stack of which customers can pick and choose which components they need to compliment their existing infrastructure or it can be an entire stack purchased as a single SKU. Deploying CloudSystem does not require you to discard your current infrastructure.
As HP is approaching converged infrastructure like CloudSystem, it is building these solutions out on industry standard (tried and true) hardware with millions of operational hours already logged on this hardware. In this case, CloudSystem is an extension built on VirtualSystem which in turn uses BladeSystem as a basis. So, although these are new offerings, they are not truly new. CloudSystem Matrix is the software used for orchestration and automation in the CloudSystem solution. This software was formerly known as BladeSystem Matrix, but rebranded to CloudSystem as the overall solution was assembled.
For storage, users can select either a P4000 (Lefthand) iSCSI array or a 3PAR array as the basis for CloudSystem. Both are modern, fresh approaches to storage and both were designed to handle the unpredictable workloads of virtualized systems.
HP is still working on CloudSystem API’s which will interact with other public clouds to allow bursting and hybrid hosting, so that work is to be released, but the vision is to be able to host with a VMware vCloud provider or any other cloud provider. They are really seeking to make this solution work with as many other cloud solutions as they can, which is what I would want as a customer. HP is working hard to avoid single vendor lock-in, in some ways hedging bets on which hypervisor will ultimately win the war.
Finally, and most importantly in a service provider’s case, HP is offering solutions were they can host a public cloud for you in a turnkey solution, where a service provider can begin marketing and selling the solution while having it hosted in HP’s datacenter. HP can provision your public cloud and hand over the keys for you to use in one scenario.
Disaster recovery (DR) is another scenario where HP’s hosted public cloud might make sense. In my company’s case, we could bring a Service Provider edition of CloudSystem onsite and offer it as a public cloud to our customers utilizing our high speed network connections in our service area, but we could also replicate the cloud data to an HP datacenter as DR solution. Since we are in a hurricane prone area, this is an appealing offer where we could assure our customers of DR failover to another geographical area.
All in all, I stand by my assessment that this is the most complete solution that I have seen to date. I know that VMware has been working very hard to build out the entire stack and vCloud Director has certainly made big steps towards the service model, but VMware is still working on issues such as VM data transit to the public cloud. HP will likely have the same issues in moving large VM hard disk images from on-premise to the cloud. This pain point can be lessened when using a local provider with lots of bandwidth available to you, such as the case with my company. This is still a fast changing area, but I am impressed by what HP introduced last week.
In the interest of full disclosure, HP and Ivy Worldwide invited me and paid for my trip to HP Discover. Even though, I am trying to relay the information as impartially as possible.