Controlling Virtualized SAP Landscapes

Virtualization of server and storage landscapes has become a common means for SAP customers to increase the efficiency of IT resources and operations while decreasing overall total cost of ownership (TCO). Allowing high flexibility in resource deployment, virtualization also offers ideal support for orchestrating composite applications within SAP enterprise service-oriented architecture (enterprise SOA).
Along with virtualization comes a fluctuating deployment of IT resources over time. This requires an advanced monitoring environment which enables auditing, problem determination, and pro-active systems management of dynamic SAP infrastructure components. SAP and IBM have co-developed a generic solution for the SAP monitoring infrastructure, enhancing today’s CCMS implementation.
To achieve this, new metrics were introduced to the SAP monitoring infrastructure that has so far been built for static environments. The new metrics – for example the capacity consumed by an logical server partition (LPAR) within a shared CPU pool –, enable dynamic virtual computing resources to be linked to the physical server they are hosted on. Both real-time monitoring and recording of history data give customers a comprehensive view of their virtualized SAP infrastructure.

Fundamentals of monitoring static . . .

The current SAP monitoring infrastructure is perfectly adapted to dedicated server environments with a fixed 1:1 relation between SAP instance and server. The SAP monitoring tools are also suitable for a virtualized infrastructure as long as the resources associated with an SAP instance stay constant over application runtime. This includes common hardware architectures where the LPARs require a reboot in order to change the size of the system. In other words, the operating system collector (SAPOSCOL) and the CCMS do not distinguish between monitoring a discrete server or a logical partition.
Even though LPARs have become dynamic with IBM POWER4 – that is, the number of CPU cores can be changed without rebooting and restarting SAP instances – the basic SAP monitoring component SAPOSCOL is able to recognize and handle these resource changes. However, although virtualization is becoming increasingly dynamic and allows for a finer granularity than physical CPU cores, SAP monitoring tools have not been enabled to cope with rapidly changing resources.

. . . and dynamic SAP environments

Based on its long experience with virtualization on the mainframe, IBM made highly flexible and dynamic concepts available for the midrange server market by introducing IBM Advanced POWER Virtualization (APV). Both virtualization technologies are characterized by a resource scheduler which supports re-distribution of compute resources among LPARs running different (SAP) loads within milliseconds. Optimal resource sharing and a minimal overhead are achieved by supporting the scheduler with special hardware instructions. Defining individual rule sets for resource distribution among multiple production and non-production SAP instances, customers will achieve a much higher overall server utilization through autonomic peak load compensation while gaining a high degree of flexibility in tailoring the virtual resources to real SAP application demands.
Two sets of parameters are of special interest for maintaining and optimizing virtualized system environments: On the one hand, parameters defining the current configuration of the virtualized system and, on the other hand, performance metrics showing its current status. The configuration parameters can be modified while the virtual systems (LPARs) and SAP applications stay active by influencing the resources inside an instance and/or re-distributing the resource shares among several instances within a resource pool. Typical adjustment parameters are:

  • the number of both physical and virtual CPUs
  • the SMT mode (simultaneous multithreading)
  • the CPU entitlement, that is, the CPU capacity that is guaranteed to the LPAR
  • the capping mode, which defines whether an LPAR can exceed the guaranteed CPU capacity as long as the shared processor pool provides spare resources.
Virtualization metrics
Virtualization metrics

On the monitoring side, several measures describe the actual resource consumption of an LPAR within the boundaries specified using the parameters above. Important values indicating the healthiness of a virtualized landscape are:

  • the currently consumed fraction of the configured entitlement
  • the remaining CPU capacity of the shared pool
  • the fraction of the shared pool capacity currently used by the LPAR.

Depending on the virtualization technology, naming may be different and additional or other metrics may be required to fully describe a virtualization setup. Reflecting such differences in metrics and terminology, an open and platform independent monitoring architecture is key for customers in order to manage application performance and optimize utilization of their hardware landscape. This results in an improved return on investment (ROI) for IT investment and faster, more pro-active operation of virtualized landscapes.

Generic implementation of the new CCMS solution

New transaction OS07N
New transaction OS07N

CCMS is the generic infrastructure for monitoring SAP applications and their underlying operating system. As a first step the data collector (SAPOSCOL) has to be extended in such a way as to provide the additional configuration and metric information used with virtualized systems to the CCMS.

SAP NetWeaver Administrator Console
SAP NetWeaver Administrator Console

A new ABAP transaction OS07N displays the extended set of data in a complete and well arranged design within the familiar user interface. The new transaction dynamically adapts to the underlying virtualization technology and the desired metrics to be displayed. The seamless integration into the CCMS infrastructure enables presentation of the additional data within SAP NetWeaver Administrator without additional development effort.
In addition to the enhancements in real-time monitoring, the complete set of virtualization-specific configuration and measurement data is stored in the Central Performance History (CPH) of the monitoring architecture. Together with other system management related information, these data can be forwarded to a BI system for central reporting. Standard CCMS capabilities can be used – for example, the operator in charge can be notified of a critical situation via SMS or e-mail whenever a configurable threshold is met. CCMS agents that provide monitoring data from remote SAP systems have been extended accordingly, supporting an easy-to-deploy central monitoring system for distributed virtualized system landscapes.
The solution allows customers to track and analyze the effects of manual resource changes on SAP systems. The history data can be mapped against objectives of the customer’s service level agreement or be the basis for further analysis, for example accounting and billing. The architecture is open for third party access for such purposes. The solution provides remotely callable interfaces to access the classical and new CCMS metrics, including their history. These interfaces either access CCMS metrics directly via BAPI calls (Business Application Programming Interface) or connect to SAP NetWeaver Business Intelligence (SAP NetWeaver BI) to retrieve aggregated history data. SAP internal and external teams are currently designing and programming against those interfaces and the first tools are being made available right now. The IBM Insight tool for SAP system capacity planning is one example of this type of application.
The first implementation of the new monitoring environment was realized on IBM POWER5-based IBM System p and System i servers. The open architecture allows easy adaptation for similar virtualization environments. This also holds true when using server provisioning instead of server virtualization – that is, a fluctuation of discrete systems compared to fractions of a server. From a monitoring perspective, provisioned environments behave very similarly to virtualized systems. The basic differences are the chunks of resources and how often resources are dedicated to SAP instances. The new solution in CCMS is therefore also applicable to installations using provisioning technologies.

Initial customer experiences show the benefits

To prove the practical feasibility of the improved monitoring solution, it has been implemented at selected customers. In cases where centralized SAP monitoring was already established, implementation was straightforward and took no more than an hour. The major benefits seen by customers are:

  • good integration into an existing CCMS environment and identical “look & feel”
  • replacement or complement for existing self-written OS based utilities (for example, UNIX sar) for monitoring virtualization
  • re-use of history data, for example for metering and billing
  • IT capacity planning and load prognosis for an entire virtualized server landscape are put on a solid calculation base.

The solution described is the first implementation of an extended SAP monitoring environment coping with highly dynamic virtualized compute resources. It provides well integrated monitoring of SAP environments running on top of IBM Advanced POWER Virtualization. The openness of the architecture allows third party tools – including customer reporting – to make use of these data as well. The solution design and defined metrics offer an excellent base for other platform and virtualization providers to develop comparable solutions.
Plans for future developments include monitoring of memory dynamics and putting the monitoring of LPARs that run SAP applications into context with overall server utilization, including non-SAP workload. Established SAP analyzing tools are planned to make use of the improved data sets, thus delivering more reliable SAP system assessments.

Olaf Rutz

Dr. Georg Leffers

Matthias Koechl