Inchoate Curmudgeon

An elaboration of the concept of an application running on a server requires a decomposition of the host type. Hosts and applications may run on a simple host where there is no IP associated with the application, applications may run in a cluster as a service with their own distinct IP, and applications may run in a virtual host environment like VMware or a Solaris 10 container or zone. For the “simple host” type, the host may be free-standing, or part of a larger complex, such as a domain or some partition of a larger frame. These concepts are elaborated only insofar as they are interesting for selection in ranking exercises. Clustering and virtualization both have a rich set of implementations and features that may not require modeling for ranking exercises.

Host Service

An operating system instance in a unique security domain or context and unique configuration elements. This may be an actual physical instance of a Server, or it may be a Virtual Host.

Application Instance

An instance of a particular an Application on a Host Service.

Virtual Host

A simple “virtual IP” or “service IP” on a server, A VMware guest operating system instance, a Solaris 10 zone or container, or a cluster resource or service (Cluster Virtual Host).

Server

An actual physical server that may be a free-standing Simple Server, or a Partition that is subset of resources from a frame or other larger server resource pool. In Sun technology, an E25K would be a Frame Controller and a domain on the E25K frame would be a Partition.

Server Virtual Host

A simple separate IP and/or port on a server bound to a particular application, a guest operating system instance (e.g. VMware guest) or a separate zone within an operating system instance (e.g. Solaris 10 zone). A Server Virtual Host belongs to a Server. Note that by this definition, an instance of a web server on a machine with a service IP separate, unique and distinct from the underlying host would be a Server Virtual Host. This is interesting and distinct from the configuration where a web server service shares the IP of the host in that it is more easily moved or relocated from host to host in consolidation exercises.

Cluster Virtual Host

A Cluster Virtual Host is a managed service that hosts applications on available physical servers within a Cluster, usually migrating in the event of a hardware failure. Generally, a given Application Instance will be hosted on a single node within a Cluster at any given time. Most clusters also have the concept os a “load balanced” service in addition to the usual configuration wherein incoming requests for a service are routed to any available server in the cluster running an instance of the application. That is, the concept of Cluster Virtual Host could also apply to a load-balanced web-farm. The term “cluster” is definitely overloaded here.

From the viewpoint of optimization and consolidation exercises, how these situations are treated will depend on the policies in place for managing the services in the cluster. For example, there will be differing potential consolidation scenarios if there is a rule or policy that any single server must have sufficient capacity to handle all loads on a cluster in the event of multiple server failures versus a rule or policy that only single-server failures must be accommodated in capacity planning.

For this reason, the important attributes and concepts for this exercise are:

• Which applications instances are in which cluster?
• What is the method for calculating “utilization” for each application instance in the cluster?

The actual assignment of each instance to a particular cluster node is irrelevant. The important thing is understanding how the cluster as a whole is under or over utilized keeping in mind compliance with availability objectives.

A two-node cluster with servers running at 50% individual host utilization each is likely at 100% cluster utilization per most “normal” policies. That is, we are more likely concerned with overall cluster capacity and utilization than simple actual CPU utilization on each node in the cluster. These considerations would also prevail for a “cluster” of web servers behind a load balancer: how much capacity is required for the entire group of servers?