OSL UVE - Virtualisation³

Hyper-Convergent Infrastructure

The OSL Unified Virtualisation Environment (UVE) combines the three technologies required for the non-stop operation of the virtual machines: server virtualisation, storage virtualisation and network virtualisation. For the first time the established standards in the OS and network virtualisation are integrated into a single solution with the unique OSL technology for the clustering, storage virtualisation and network-I/O. This way a completely virtualized IT infrastructure module was created, which itself redundant, flexible and scalable, and can add services to the existing IT infrastructures at defined transfer points.

Distinguishing features:

  • radically simplified infrastructure - everything over one physical network
  • no hypervisor-specific knowledge required
  • infrastructure scalable in all directions
  • integrating various storage systems
  • no dependandy on a specific hardware manufacturer - realization on the hardware of your choice

OSL has been consequently focusing on the complete process of the delivery, operation and maintenance of the high available, dynamic infrastructures, and with the UVE a solution package was created, which enables enormous simplification and reduction of costs.
Dividing of the responsibilities between the storage / SAN / network / server and OS administration, the application management, operation, backup and recovery is nowadays oftentimes a preferred way to deal with the enormous complexity of the dynamic, virtualized and high available infrastructures. This is often reflected in the division of the total process which is unified in itself into multilpe departments. At the places where the products of different manufacturers with oftentimes complex concepts has been established for solving of only parts of the whole mission, we can often observe interface issues, functional losses, additional time, energy and costs involved, and also insecurities which can become serious challenges even for big data center operators, and for the smaller users it would oftentimes completely ban the access to an adequate and affordable complete solution. Here the OSL UVE offers a complete solution.

 

 

Standard UVE

Standard UVE

The data center owners who are already operating a classical data center infrastracture with RAID systems and SAN, can continue using them for their VM infrastructure. The UVS servers take their storage resources from the SAN and provide them as virtualized resources to the VMs. At the same time, the UVS also provides other necessary functions, like Virtual Network, VM Management etc. The illustration shows a redundant layout with 2 UVS.
The good part: the UVCs, i.e. the hosts on which the VMs are running, are connected redundantly over Ethernet (or IB) only. The network is convergent and transports both the regular LAN traffic of the VMs and the I/O requests. This enormously simplifies the configuration, eliminates potential sources of errors and saves relatively expensive and unflexible SAN ports. The Block I/O protocol RSIO developed by OSL guarantees the best performance, scalability and redundancy.


Slim UVE

Slim UVE

If you do not want to invest anymore in the classical data center stack, separate RAID systems and FC, or if you are looking for the highest possible performance, then the Slim UVE is a good choice for you. Thanks to the internal SSD/NVMe disks used in the UVS, the UVCs have a very short path to their storage resources. The illustration shows a simple configuration where the UVS is "taking the role" of a RAID system, but with highest possible performance and all additional functionality necessary for the operation of a VM infrastracture (VM Management, HA Management, Storage Virtualisation, Virtual Networking etc.)


Slim UVE redundant

Slim UVE with redundancy

The "Slim UVE" can be clustered, too. In this case, the loss of one UVS can be bypassed without downtime for the UVCs / VMs.

Supported hardware

Servers

The Unified Virtualisation Environment 4.8 is a highly portable HCI with a client-server architecture. The UVE can in principle run on Solaris and Linux (also mixed). As a rule, Solaris systems (SPARC or x86) or Linux systems are used for the UVS, while for the UVC almost always Linux-based x86 systems are used. The choice depends on the existing data center environment and on the customer-specific decision criteria. Please follow the OSL recommendations when choosing the operating system (see software descriptions / data sheets). The available options for the server hardware are resulting from the chosen operating systems. In particular, the OS choice determines which network, SAN or storage controllers can be used.

Unified Virtualisation Clients (UVC) build the Compute Node Farm and act as Hypervisor Nodes. The Virtual Machines run on the UVCs. By adding more UVCs the system can increase the computing capacity, performance and availability both in computing and in I/O. This requires CPUs with virtualisation extensions. As for the operating systems, both Solaris and Linux are possible in principle, usually Linux systems (x86) are used.


Operating Systems

  • Linux SLES 15.5 (Tumbleweed or other systems upon request)
  • Solaris 10 / 11, recommended for UVS 11.3, for UVC 11.3/11.4

Please also consider the release notes of the manufacturers for the related systems.


Network

The Unified Virtualisation Network (UVN) connects the UVS to the UVCs. It is recommended to use a multipath connection between the UVS and the UVC. Depending on the configuration, the Slim UVE can be reliably implemented even using only one switch and internal SSD. For the Unified Virtualisation Network at least 1G-Ethernet technology should be used. For ultra-short latencies, the IB or at least 25G-Ethernet, better 100G or higher should be preferred.


Shared Storage / RAID Systems

Shared Storage / external RAID systems can be used in UVE. In this case attention must be payed to the appropriate (supported) I/O controllers on the UVS. However, thanks to the storage virtualisation capability integrated in the UVE, using external RAID systems is not necessarily required, because the replication between the two UVS is possible through the engine itself. This might be advantageous from the costs and performance point of view. If only one UVS is used (or also in some other situations), the usage of an internal RAID controller might be of advantage. Please discuss these questions with your system house or with OSL.

 

Administration of the virtual data center from a single chosen node

The OSL UVE can be scaled to a large number of the clustered nodes and virtual machines. The cluster stack is completely integrated and is delivered with a standard installation. In order to deal with all the tasks of your virtual data center, you can use the command-line interface or the web-based graphical user interface.

Command line
The complete management and operating of the virtual data center can be done on the command line. For this, the documentation in form of manual pages is provided. With the help of the command line you can administer your whole cluster from any UVS node. You do not need a dedicated manager node for this.

WebGUI - web-based graphical user interface
The graphical user interface of the UVE is integrated with the cluster and provides a comfortable way to control your virtualized data center. The web-based interface can be used with a regular browser (recommended Firefox). The WebGUI helps you to operate your cluster and to observe the state of every single node or VM. The WebGUI can be also used to manage the storage resources and the cluster VMs. Furthermore, an extensive performance monitoring of the cluster nodes and the guest systems is available as a live log and also as a history log. The data for the live log (CPU, memory, I/O, network) are available for the last hour with the 4-seconds intervals. The history log is persisted in a file system in a special storage-saving binary format. It comprises the minimal, maximal and average performance data with the 1-minute intervals. This way you as a system administrator can be always informed of how much stress is on your systems.

An administrator can give further persons a restricted access to the WebGUI. For this purpose the GUI incorporates a permission-based user management integrated with a system of the so called Organisation Entities:
An Organisations Entity (OE) depicts an element in the organisational structure of your enterprise. An OE can represent a department, a group etc. In the UVE the OEs can be managed in a hierarchical structure with up to 3 levels. Roles are composed of permissions and grant access to specific resources or actions (e.g. VM start/stop) in the UVE to specific users. Also the access to the specific areas of the WebGUI depends on the permissions. A user can have one or more roles, while every user role can be restricted to specific OEs.

UVE-WebGUI

Supported Virtualisation Technologies

Linux is currently the preferred system for the hypervisor nodes (UVC) in the OSL UVE. The following virtualisation mechanisms for these hypervisor nodes are supported:

- KVM
KVM is the leading virtualisation technology for the full virtualisation on Linux. It is integrated as a kernel module into the Mainline Linux Kernel and runs with nearly native performance on any x86-hardware with virtualisation support - either Intel VT-x or AMD-V.

With KVM you can run both Windows and Linux in virtual machines (VMs) ausführen, while every VM has its own virtualised hardware: a network card, a storage disk, a graphic card etc. Since you can run multiple applications in VMs on a single piece of hardware allows you to save the power and costs, and at the same time you have the flexibility to establish an agile and scalable software-defined data center, fitting your business requirements.

- VirtualBox
Storage disks are emulated in container files, also called by VirtualBox as Virtual Disk Images, (kurz VDI). Apart of this own file format VirtualBox can also deal with the HDD files of Parallels and with the images in the formats QED (QEMU enhanced disk) and QCOW (QEMU Copy-On-Write) of the emulation and virtualisation software QEMU. Apart of that the iSCSI can be used as virtual storage disks, while the required iSCSI initiator is already contained in the VirtualBox. With the VBoxManager command-line tool belonging to the VirtualBox also these foreign formats can be converted.
It is possible to operate VirtuaBox with further guest operating systems. The activation of the virtualisation extenstion of the modern x86 processors (Intel: VT-x, AMD: AMD-V) can help to run an otherwise non-supported OS in the virtual environment of the VirtualBox.

Block I/O and IP over one network - simplified infrastructure

The RSIO protocol integrated in the UVE allows the hypervisor nodes to have the block I/O access to a global virtualised storage pool over network. The same physical network is divided between separata clients and used for the data flow of the virtual machines and for the administration of the cluster services. This results into a radically simplified infrastructure, making multiple administrative tasks unnecessary, which also means more clarity and an improved protection from the failures. In principle, there is no need anymore for LUN masking, administration of the FC fabrics and networks switches, multipath configurations for the block I/O, complex network configurations ans IP multipath solutions in VM guest systems, the usage of VLANs in virtual machines and an extensive device managements for virtual machines.

RSIO-Architektur

Besides well-organized infrastructure which is easy to overview, the UVE network solution can also have substantial cost-saving effects, caused for instance by not having to use the FC fabrics and reducing the number of used network ports. Further positive sides of this implementation are high port density, lower energy consumption, and reduced space requirements.

Availability and scalability on all levels

With the OSL UVE all the levels of the complete solution can be decoupled in the best way possible and set-up redundantly. Unnecessary redundancy can be spared, the complete solution stays simple and easy to use, and literally all hardware components can be replaced without service interruption.

The intelligence and responsibility for managing all the substantial functions are moved from the partial components to the UVE server (UVS). This helps to reduce the complexity and interdependencies, improves the availability and allows the online replacement of single components, also by products of different manufacturers with other performance and configuration parameters and other features. Therefore the older RAID systems can be replaced for example by a newer generation, the switches or the hardware of the hypervisor nodes or even of the UVS can be exchanged without having to interrupt the cluster operation.

UVE-Redundanz

High Availability

We at the OSL go with our products alternative ways here. The high availability is ensured not over an extra cluster interconnect or a quorum device, but over an automatic self-management of the applications.

Many data center applications should stay online without interrupts for business administration reasons. To fulfill this requirement, an efficient protection from hardware failures is needed. Cluster solution can grant such protection. In case of a disaster, they are responsible for migrating applications onto alternative hardware. Another suitable host takes over the application in such cases. Offline times are avoided. This procedure, however, requires an abstract access to the hardware resources. Processes and applications must be described and adjusted if necessary.

Active/active, active/passive or none of them?
Conventional cluster solutions are known to be either active/active or active/passive. The active/active solutions require at least two application servers to be online, and in case of the failure of one server, the second server takes over the duties of the first. The active/passive solutions require one standby-device to be online, which can take over the job of a primary system in case of its failure. Although also such architectures are supported, the OSL cluster concept is not limited to these two traditional approaches. The OSL UVE can also represent much more intelligent configurations. Our software controls your cluster applications automatically via resource-based application self-management and assigns them to the hosts that are available according to their resource requirements. In cases of hardware failures the affected applications are reordered and assigned to the remaining hosts according to their priorities:

The OSL user assigns priorities to the applications. Based on the performance benchmarks, the application descriptions and priorities, the applications get assigned to different nodes in the cluster.  This can happen automatically depending on the capacity of the available hardware.

If a cluster node experiences a failure, the cluster launches the process of the redistribution of the applications over the remaining cluster nodes. Applications with higher priorities can even drive away the applications with lower priorities, so that the applications with higher priorities can stay online. This way the user can assign higher priorities to the production applications, and lower priorities to the test, development or QA applications, so that the important applications stay available even in case of disaster.

If the failed hardware later becomes online again, the stopped applications can be restarted. Another migration of the running applications is not necessarily required then.