Cluster systems

As a cloud provider, Parking.ru has experience in providing not only services with “normal” reliability, but also high-availability hosting services built on a clustered, duplicated infrastructure.

Having significant experience in building such infrastructure projects, we decided to offer it not only to selected clients (who have been using these services for many years), but to make a standardized offer for everyone.

Parking.ru launched a number of cluster solutions based on a reliable duplicate infrastructure and based on a variety of solutions: from virtual machines to a group of physical servers. Separate attention is given to the proposal for GEO clustering in two different data centers combined by duplicate communication channels.


')
If we consider briefly, then within the proposal there are the following typical cluster schemes:

High Availability Clusters

Created to ensure high availability of the services provided by the cluster. Excessive number of nodes in the cluster, ensures the provision of service in case of failure of one or more servers. The minimum number of nodes is two, but maybe more.

Typically, a High Availability cluster is built for Microsoft SQL servers, which support Internet project databases. A high availability cluster can also be built for Exchange systems.

There is another type of High Availability SQL Cluster - “database mirroring” or database mirroring. This type of cluster does not require dedicated disk storage, but for automatic switching in the event of a crash, another SQL server is needed - witness. Such a cluster is ideally suited for WEB applications and requires less creation costs.

Load Balancing (Network Load Balancing, NLB)

The principle of operation of NLB clusters is the distribution of incoming requests to several physical or virtual server nodes. The initial goal of such a cluster is performance, however, they are also used to increase reliability, since the failure of one node will simply lead to a uniform increase in the load on the other nodes. A cluster of cluster nodes is often called a cluster farm. The minimum number of nodes in a farm is two. Maximum - 32.

When hosting high-load web projects in NLB mode, a farm of web servers is built on IIS 7.x

Clusters on virtual machines

The most affordable and scalable solution is to build a cluster based on virtual machines on the Hyper-V platform.

Virtual machines with Windows Web Server 2008 installed on them (IIS7.x, user application) are used as NLB cluster web boxes.

As a database cluster, two virtual machines of required capacity are used on Windows Server 2008 Standard Edition and SQL Server 2008 Standard Edition.

Fault-tolerant unified data storage based on Storage System from NetApp or HP.

To ensure greater reliability, all cluster nodes are located on different physical servers / cluster nodes.

The scalability of the solution is achieved by increasing the power of the virtual machines used (up to 100% of the physical server capacity), as well as by adding new nodes to the NLB cluster.

With the use of online migration tools, over time, it is possible to transfer cluster nodes to new, more modern physical servers without loss of performance and without downtime of any node.

This cluster is the best solution in the price / quality ratio and is recommended for both business-critical applications and relatively loaded web-projects (~ up to 30000-50000 visitors daily).

Clusters on physical servers

Web projects with a load of over 50,000 visitors, projects with special security requirements require building cluster solutions on dedicated servers without loss of physical server capacity for virtualization.

The scheme for building such clusters is similar to the scheme for building clusters on virtual machines, only dedicated physical servers are used as nodes.

Geoclusters

When building local clusters, a single point of system failure is the data center infrastructure itself. Parking.ru offers a unique solution on the Russian market - building a geographically distributed cluster, the nodes of which are located in different Parking.ru data centers.

Each of the cluster nodes has independent access to the Internet, but at the same time, the cluster from the data network looks like a single server with a single address and content.