As well as why Absolute of Insurance refused Oracle VM and transition to "clouds"

For the end of 2014 the Absolute Insurance company had about 12 physical servers, on 4 of which the system of virtualization Oracle VM which allowed to place about 10-15 virtual ^[1] was installed^[2].

Oracle VM is the system of virtualization from Oracle constructed based on Xen. As the operating system of domain 0 can be used Oracle Enterprise Linux or Red Hat Enterprise Linux. In the guest systems the same systems, as in normal Xen can work. Are a part of Oracle VM: Oracle VM Manager and Oracle VM Server.

Xen is the cross-platform hypervisor extended on the terms of the license GPL. Main features: support of the mode of paravirtualization (in addition to hardware virtualization), minimality of the code of the hypervisor due to carrying out of the maximum quantity of components out of hypervisor limits.

Paravirtualization — technology of virtualization at which guest operating systems are prepared for execution in the virtualized environment for what their core is slightly modified. The operating system interacts with the program of a hypervisor which provides it guest API, instead of use directly of such resources as the table of pages of memory. The code concerning virtualization is localized directly in the operating system. Paravirtualization requires that the guest operating system was changed for a hypervisor, and it is a lack of this method as similar change is possible only if guest OS have open source codes which can be modified according to the license. At the same time paravirtualization offers performance almost like at the real not virtualized system and also a possibility of simultaneous support of different operating systems, as well as at complete virtualization.

Failure from Oracle VM 2

Oracle VM minuses:

• Updating of Oracle VM from version 2 to version 3 required reinstallation of all servers as in the last system architecture was considerably changed;
• The policy of Oracle company "left much to be desired", a system could become paid or cease to be supported that already happened to some products of the company;
• Rare updates of a system;
• Excess resource consumption by Xen hypervisor, impossibility of increase in resources of the virtualized system to "raid";
• Existence of the client-server architecture: so, at failure the Oracle VM Manager servers all system need to be installed almost from scratch.

Possible versions of solutions:

• Transition to Oracle VM 3 (option, not suitable for our company, considering the above described minuses);
• Transition to "clouds";
• Transition to more lightweight and less resource-intensive solution, without the client-server architecture.

Reasons for refusal of transition to "clouds"

1. High monthly costs for placement of servers in "clouds".

Value assessment of transition to "clouds" was carried out at the end of 2014. To "clouds" it was going to transfer the main servers of the company:

• Exchange server,
• server 1C,
• server of Oracle.

Having studied offers of the known companies providing services in placement of servers in "clouds" we found out that the average monthly payment for placement of 3 servers makes about 120-160 thousand rubles. We considered similar costs inexpedient.

Also for comfortable work of our employees (both head office, and branches) with the servers placed in "clouds" the company would be forced to increase considerably capacity of Internet channel that is expensive "pleasure" for the company located in business center with provider monopolist.

2. Prohibition on placement of servers not in the territory of the company.

Thus, the option of transition to "clouds" was impossible in our case. There was a need of search of other solution of a problem. Having carried out the analysis of the available systems of virtualization, we came to a conclusion that use of Proxmox Virtual Environment will be the most appropriate solution.

Transition to Proxmox VE 4.x + Ceph

It was decided to pass to a linking of Proxmox VE 4.x + Ceph that would allow to create scalable and failsafe own "cloud".

Ceph is free program object network of storage (engl. object storage) providing both file, and block interfaces of access. It can be used on systems consisting both of several Linux machines and from thousands of nodes, it is known of operation of systems on Ceph of hundreds petabyte in size (in Yahoo!); the built-in mechanisms of the duplicated replication of data provide high survivability of a system, during the adding or removal of new nodes the data array will automatically be rebalanced taking into account changes.

Unlike such distributed file systems as GFS, OCFS2 and GPFS, in Ceph data processing and metadata is separated into different groups of nodes in a cluster, approximately as it is made in Lustre, with that distinction that processing is made at the level of the user, without demanding any special support from a kernel of operating systems of nodes. Ceph can work over block devices, in one file or using the existing file system of a node (for example, XFS).

In Proxmox VE 4.0 OpenVZ it was replaced with LXC. When converting images of machines from OpenVZ in LXC observed some problems, for example:

• The container LXC works by 2 - 3 times more slowly than OpenVZ;
• Containers with Debian 6 and CentOS 5 are not supported;
• In CentOS 6 the network because of renaming of network interfaces did not work with venet0 on eth0;
• It is impossible to come into the container redhat based on ssh because of lack of tty;

In version 4.3 all corrected.

For the end of 2016 there is a creation of the test stand for implementation of own "cloud".

Notes