FIS-ASP, together with Suse and the server storage specialist Thomas-Krenn AG, has created a cloud-based data center infrastructure for SAP medium-sized businesses. It is based on OpenStack and is optimized for short provision times.
Medium-sized SAP users are also becoming increasingly international, which is why the requirements for hosting are rising. On the one hand, the provision time becomes a critical success factor. New tasks also arise from “SAP HANA” or from “S/4” projects, migrations, “SAP EWM” implementations….. While a three-level SAP system landscape used to be sufficient in the past, other landscape components, such as SAP Fiori Frontend Servers, Adobe Document Services or Web Dispatchers, are often used today.
The challenge lies in combining the systems with each other and with existing landscapes. In addition, updates and changeovers must often be possible in parallel. Therefore FIS-ASP, a subsidiary of the FIS Group and specialized in the implementation, operation and maintenance of global SAP landscapes, launched a project in 2013 that dealt with the OpenStack technology still young then. The aim was to use the technology to create an automated cloud solution for the SAP landscapes hosted by FIS-ASP.
OpenStack allows the desired automation
Suse GmbH was a partner from the very beginning. FIS-ASP now operates almost three quarters of the SAP systems on Suse’s “Linux Enterprise Server”. OpenStack as a project supported by Suse for the development of a free architecture for cloud computing enables exactly the desired automation.
After one day and by using two compute nodes, a controller and an admin node during the proof of concept, FIS-ASP achieved the same result as it had achieved in the previous tests only after several and was therefore able to directly start creating and preparing images for the SAP Basis installation. With the “Crowbar” module, Suse integrated a software solution for automated “fueling” of the physical nodes of a cloud.
In the course of the tests, FIS-ASP switched from SAN for the operating system on this node to local disks. An SLES operating system with the default KVM hypervisor was rolled out to the compute node.
But: The use of SAN storage for the host operating system still presents only problems with “fueling” and does not bring any added value in an environment where only a hypervisor with an additional OpenStack “Nova” service runs on the compute node. Nova is the actual core component of OpenStack, which is responsible for starting, stopping and managing virtual machines.
Controller node in highly available cluster variant
Another technical milestone was the transition of the controller node, which, up to now, had only been installed as a single server, to a high-availability cluster variant as delivered in the Suse OpenStack version. All critical OpenStack components are set up on a new “Pacemaker” cluster as a two -, or even better, three-node variant with high availability. Critical components include a central database with “Postgress” and a fast message service based on “Rabbit MQ”.
FIS-ASP used update windows to a new version of “Suse OpenStack Cloud” to successfully perform this changeover with a necessary reinstallation of these core components on a cluster
The provider network on VLAN/VxLAN keeps cloud architecture simple
Another important factor during the setup was the selection of the right network setups. The “provider” network available in the OpenStack makes it possible to build on proven VLAN/VxLAN technology and to keep the cloud architecture intentionally simple but more robust in operation.
The number of compute nodes in OpenStack can now be easily scaled upwards and is complemented by a series of features such as availability zones and high availability of compute nodes. Behind the availability zones, fire compartments in the data center or remote data centers can be defined where the virtual machines or landscapes have to run.
In the latest version of Suse OpenStack Cloud, the high availability of compute nodes or specific workloads has been added – an essential prerequisite for the operation of business-critical systems in the cloud. This was made possible by using the “Pacemaker Remote” technology, which uses the existing cluster of the controller pair.
The only thing missing was a shared storage to keep the systems as available and efficient as in a traditional virtualized environment. Various SAN and NAS providers were selected and tested. Especially in the beginning, there were incompatibilities between the available OpenStack-related “Cinder” drivers and the “OpenStack Storage API”. As a central OpenStack component, Cinder manages block storage and its use.
Ceph ass shared storage
The solution was found with “Ceph” (again in the OpenStack environment), one of the most widely used software-defined storage solutions for OpenStack on the market. Ceph is highly scalable, runs on the basis of any standard hardware and has been implemented as “Suse Enterprise Storage” standard software subscription similar to the cloud software.
The basis for FIS-ASP was a reference architecture from Suse GmbH and Thomas-Krenn AG, whose hardware is perfect for building a Linux-based ceph storage system. Two NVME disks are installed in each of the six nodes to ensure distributed access of databases with high performance. NVME SSDs are used for journaling and are the ideal medium for this purpose: They have a high performance, can handle many permanent write accesses and have an extremely long lifetime.
Images, automation, standardization
The general goal of cloud systems should be to always use the latest versions, especially for security reasons. However, not every combination of system, database, operating system and different versions makes sense in terms of security, stability, performance and cost. Real added value can only be achieved with the right combination of standardization, automation and a functioning image and version database.
Here, OpenStack and Ceph show their strenghts with their open architecture as open source software. In addition, FIS-ASP uses further open source applications such as the “Github” software for version management of the correct images, “Salt” for automating post-processing and “Heat” templates for building virtual machine images – all coordinated and flexible enough for the tough requirements of SAP operations.
Here, it is important that the SAP system ID can be selected as well as the customer domain or the instance numbers used. In OpenStack, customers are created as ‘tennants’ and supplied with predefined rules via the integrated firewall. In the last two years, SAP has stabilized and better documented the API of its “SAP Software Provisioning Manager”. The SWPM is used to install the SAP systems in order to implement the above-mentioned options (SAP SID, domain, instance number…). This means that many installations can now be selected and installed using templates.
, Head of Delivery SAP at FIS-ASP GmbH, created a service offer for medium-sized companies together with partners from Suse and Thomas-Krenn AG as well as with SAP and open-stack software. (Picture: FIS-ASP (Bild: FIS-ASP)
The FIS-ASP offer
Based on this cloud infrastructure, FIS-ASP is able to offer its customers dedicated services in the areas of Infrastructure-as-a-Service, Platform-as-a-Service and Software-as-a-Service – with short implementation times for new systems and landscapes even in complex projects. In the PaaS environment, the hosting provider interconnects several SAP development systems for “SAP EWM”, “SAP Netweaver” Gateways, Web dispatcher or back end ERP (respectively S/4) into a network, for instance, where developers and customers can start with a complete development landscape in a short period of time.
Another important and growing area is “SAP Hybris Commerce”. With its setup similar to the LAMP stack setup, the shop is well suited for the use in the cloud. LAMP (acronym for “Linux, Apache, MySQL and PHP”) can be used to make dynamic websites available. In contrast to SAP Netweaver-based systems, the various SAP Hybris components including the database and application servers for front end and back end as well as other units fit very well into the cloud.
* Matthias Braun, Head of Delivery SAP, at FIS-ASP GmbH.