When enterprises consider putting business-critical workloads in public clouds, many of them overlook, at least in part, critical issues of economics and over what portion of their cloud software stack the cloud vendor has full control. This leads to a situation where sometimes relatively inexpensive offerings where the vendor has full control of their software stack (like Amazon EC2) are improperly compared to an offering from a vendor like Savvis or Terramark who is building a cloud out of either VMware-provided components or OpenStack components. This gives rise to important issues that drive both the cost of the respective offerings and the degree to which the cloud vendor can both enhance the offerings with rapid agility and quickly address service level issues.

Public Cloud Computing – Is there One Throat to Choke?

If you are going to put business- and performance-critical applications in someone’s public cloud, then you would expect to be able to hold that vendor accountable for the resulting quality of service provided as those applications are delivered to their users. In some cases, the ability to deliver quality service will be contingent upon the cloud vendor having ownership and control of the software that comprises their stack. This is certainly the case with respect to Amazon EC2 and Microsoft Azure, each of which own 100% of the software that comprises their clouds.

All other providers of public clouds rely at least in part upon layers of software from other software vendors. Google relies in part upon Red Hat KVM. Any vendor that offers up a VMware-based vCloud product relies at the minimum upon vSphere and may also rely upon VMware vCloud Director as the Cloud Management layer. Any public cloud vendor that offers up an OpenStack-based cloud will most likely be relying upon Red Hat KVM, as well as someone to provide a supported distribution of OpenStack. VirtuStream is in a unique position here, in the respect that they rely upon vSphere or PowerVM, but have their own Cloud Management layer, xStream.

So how important is this? It matters if you expect to put important workloads in that cloud, and if you expect the cloud vendor to be able to resolve issues with excellent agility and responsiveness. Due to the huge installed base of the VMware vSphere product in enterprises, it is safe to assume that public cloud vendors are not going to be dealing with legions of bugs in the underlying hypervisor. This means that any cloud vendor who relies upon vSphere needs to be evaluated on the basis of how much control they have over their cloud management layer. It would be fair to say that any cloud vendor who relies upon both KVM and OpenStack has little to no control over any of the underlying bits, and is therefore nothing more than a reseller of software that they do not control.

Cloud Vendor Hypervisor Cloud Management  Stack under Full Control of Cloud Vendor License Fees Paid to Software Vendors On Premise Version Available Quality of Service Level Guarantees
Amazon EC2 Owned by Amazon (Derived from Xen)
 Developed and owned by Amazon Yes – Amazon has full control of all source code  No Yes – but only through a separate company – Eucaplytus Abysmal to non-existent
Microsoft Azure  Owned by Microsoft (derived from Hyper-V)
 Developed and owned by Microsoft Yes – Microsoft has full control of all source code

 No

Yes – over time as Azure components get integrated into Windows Server

At the application layer through partnerships with AppDynamics and New Relic – no end-to-end service level guarantees.

Google Compute Cloud Red Hat KVM  Developed and owned by Google
 No – Google relies upon Red Hat for KVM, but has full control of its Cloud Management layer  Business relationship between Google and Red Hat is unknown
No  Unknown (service not announced yet)
VMware vCloud Offering  VMware ESXi
 VMware vCloud Director
No – Cloud vendor controls neither the hypervisor nor the Cloud Management layer  Yes – Cloud vendor has to pay usage-based licensing to VMware
Yes – VMware vSphere and vCloud Suite Dependent upon cloud vendor – but there are no known examples of end-to-end service level guarantees
OpenStack Cloud Offering Usually KVM OpenStack No – Cloud vendor controls neither the hypervisor nor the Cloud Management layer Yes – Cloud vendor has to pay for both KVM and for OpenStack Yes – through Piston Cloud, Nebula, and StackOps  Dependent upon cloud vendor – but there are no known examples of end-to-end service level guarantees
Virtustream Choice of VMware ESXi or PowerVM  Virtustream xStreme No – Virtustream relies upon either ESXi or PowerVM for the hypervisor but owns and controls its own Cloud Management layer Yes – Virtustream has to pay for the hypervisors, but not for the Cloud Management layer Yes – xStream S Yes – Virtustream uniquely offers end-to-end application response time guarantees for business critical applications like SAP

Public Cloud Computing – Vendor Economics

Amazon has clearly done an excellent job of driving down the cost of public cloud computing and establishing itself as the leader on the front of pricing and value for the dollar. This has put tremendous pressure on the rest of the public cloud computing industry. Cloud providers who license vCloud from VMware have found that they cannot offer a vCloud-based service that is price competitive with Amazon. As a customer of cloud services, this creates a clear distinction. If you want a public cloud that is based upon the same enterprise-class performance and scalability that vSphere offers on-premise, then you are going to have to pay more for that cloud than you would pay at Amazon.

The above dynamic has led many cloud providers to offer two solutions: one based upon vCloud (the VMware-compatible choice), and one based upon OpenStack (which allows them to compete on price with Amazon). This is all well and good for the cloud providers, but customers should be aware that you cannot have your cake and eat it, too. If you use OpenStack, you are not going to get the same quality of service and agility of response that you would get with a vCloud-based offering, given the diffused nature of the responsibility for the various layers of the stack that are required to construct the service.

Public Cloud Computing – Vendor Service Levels

Since this article is about putting business-critical and performance critical-applications in public clouds, the question of how to define, measure, and guarantee service levels has to be addressed. This is one area where the cheaper the cloud offering is, the worse the answer is going to be. A good answer starts with a proper definition of performance. Performance is equal to end-to-end application response time. Any service level or SLA that is not focused upon application response time is simply worthless. The fact that several Amazon-hosted properties were inaccessible during Amazon’s outage last year and that Amazon’s SLA was not violated while those properties were inaccessible simply proves this point.

This brings up an area where most public cloud offerings are extremely immature. Amazon offers no way to guarantee that latency in their infrastructure is not causing response time issues in the applications. Microsoft has application performance management partnerships with AppDynamics and New Relic, but similarly provides no mechanism to ensure that Azure itself is not the cause of the issue. VMware provides no measure of end-to-end infrastructure latency for vSphere or vCloud, so any cloud based upon its products is similarly impaired on this front. The only vendor that has effectively tackled this issue is VirtuStream, who actually measures application response time from the edge of the application system and offers SLAs guaranteeing response-time-based performance.

Summary

When it comes to public cloud computing services, that old adage of “fast, cheap, or good – pick any two” certainly hold true. Amazon can offer you cheap, and since they own their stack, a rapid cycle time DevOps approach to support. But you are not going to get enterprise-grade service level guarantees for Amazon’s pricing. To get both agile responsiveness and enterprise grade SLAs, you are going to have to give up on cheap.

Share this Article:

Share Button
Bernd Harzog (326 Posts)

Bernd Harzog is the Analyst at The Virtualization Practice for Performance and Capacity Management and IT as a Service (Private Cloud).

Bernd is also the CEO and founder of APM Experts a company that provides strategic marketing services to vendors in the virtualization performance management, and application performance management markets.

Prior to these two companies, Bernd was the CEO of RTO Software, the VP Products at Netuitive, a General Manager at Xcellenet, and Research Director for Systems Software at Gartner Group. Bernd has an MBA in Marketing from the University of Chicago.

Connect with Bernd Harzog:


Related Posts:

6 comments for “Public Cloud Computing – Economics and Throats to Choke

  1. Tom
    September 27, 2012 at 10:35 AM

    I think that we only get the agility and predictability in price we want when we give up parts of the control we’d other like. Information technology (IT) relies on trust and co-operation throughout our stacks. Just as a conventional business can only grow when its founder is able to delegate functions like bookkeeping, janitorial upkeep, retail sales, and so on, we in IT only make the most of the Cloud as we’re able to build on pieces outside our control. As with bookkeeping and other responsibilities, though, we need to be realistic about the hazards of reliance on those pieces, and put auditing systems in place to ensure the results. When a start-up founder requires his own approval of all purchases, he’s created an overload for himself; when he ignores accounting, he’s inviting embezzlement. Somewhere in the middle, there’s a healthy balance that ties trust to auditing and oversight systems.

    In a similar way, we achieve the agility and cost flexibility we want from the Cloud only when we can trust suppliers. A big part of that trust is to find the right balance. In coming weeks, we’ll look at the possibilities to keep control through performance measurement. Harzog is absolutely right to underline the importance of end-to-end results; we’ll return so on with more to say about application performance management (APM), especially in its end-to-end aspects.

  2. levine walters
    October 2, 2012 at 12:08 AM

    When price is the issue, a lot of things may suffer along the way. This is true for the switch to the cloud as well. Many things should be considered when moving to the cloud which could be of major importance than the price.

Leave a Reply

Your email address will not be published. Required fields are marked *


− 1 = six