Effectively managing application performance and availability in the cloud may seem like a simple matter of visibility and integration, but it actually stems from the relationship between application components and resources. To resolve that issue, you must first know how end users view the cloud.
End users who see the cloud as a simple tool in server consolidation view cloud management as a way to secure and activate cloud services. To users who see the cloud as a resource in an expanding web of application components, cloud management is a critical element in ensuring Quality of Experience (QoE)
Not surprisingly, it's the latter group that wrestles with management strategies for cloud applications.
QoE is the sum of the Quality of Service (QoS) among network connections in a workflow, plus the processing of the application components that the network links. In the cloud, components are assigned to resources at different locations in the resource pool. Obviously, network connections will change if an application is moved from one cloud to another or from one provider's data center to another, but other factors may influence network performance, such as congestion in the storage area network or a greater load on the servers that host the virtual machine (VM) in the new data center.
These performance issues aren't directly visible to an enterprise's management tools since they occur inside the cloud, so any information about these public resources comes from the cloud provider.
Integrate cloud management data with mPaaS
One way to integrate cloud management data with an enterprise network and IT management information is to use Management Platform as a Service (mPaaS) from the cloud operator. Essentially, mPaaS creates a management window into the cloud provider's management system. A cloud provider's integration management view links with a company's management API into a common view. This offers the best management visibility and can ensure that enterprise users and their cloud providers have a single view of cloud resources to enforce a service-level agreement (SLA).
MPaaS offers the best management visibility and can ensure that enterprise users and their cloud providers have a single view of cloud resources to enforce an SLA.
Only a few cloud operators, such as Hewlett-Packard, IBM and Microsoft, offer integrated management services, and most only do this on special order, but insiders believe more will do so in the future. For now, expect Platform as a Service providers or large IT providers to offer more capabilities here.
It may still be possible to create an integrated management view, even where mPaaS is unavailable. The question is, "How much cloud management data can be made available by the provider, and how can it be augmented as needed?"
Some cloud providers will offer management statistics on dedicated instances. The use of dedicated instances -- services that limit the geographic scope over which an application can be hosted -- will help stabilize network variables and may also reduce variations in server VM performance.
When stats on dedicated instances are unavailable, you can install some management components as part of the cloud-hosted application image. This, at least, can provide an application-eye's view of local resources and connection performance. Be sure to check with the cloud provider about management tool compatibility before installing any third-party tools.
When mPaaS won't work and full visibility into server and network connections inside the cloud isn't available, then management responsibility has to be divided among the players and meet-points must be established to permit SLA writing and enforcement. The goal is to reduce the number of jurisdictions involved in management -- and to improve network visibility. The network connects end users to the cloud, so it creates a potential management black hole. Try to get the cloud provider to take responsibility for the network connection or use virtual private network services that offer detailed monitoring and management. This will help reduce the impact of invisible network changes on performance and simplify enforcing SLA terms.
If all else fails, management boundaries are set by the connection points of the players involved in a public or hybrid cloud application. The major variable is the network, particularly when the Internet is involved.
By starting with a measurement of application response time and then subtracting measurable delays, it's easier to spot trends or changes that would indicate a QoE problem. After that, you can direct an inquiry to the entity responsible for that portion of the cloud infrastructure. This approach is often better than nothing, but most users will still yearn for widespread mPaaS.
About the author
Tom Nolle is president of CIMI Corp., a strategic consulting firm specializing in telecommunications and data communications since 1982.