The R* Plan: a network of resistant communication

Technical details

This page needs to be updated Technical Documentation: Orange Book.

With this section we would like to explain to curious people the mechanism underlying the new server network (a project known as “R* Plan” :) What we are trying to do is to describe this change in a rather easy way for those who are interested in technical issues without being supertechies.

Layer zero : hardware

The R* Plan consists in the dislocation of ‘n’ servers in ‘n’ places. The basic idea is that there is no way to prevent with certainty an unwanted physical access to the server and that it is therefore better to use instruments that can detect any unwanted intrusion and warn about any jeopardy in a node of the network.

Layer one : the network

You can imagine the network linking the servers as a net made of rings. Each ring has different access rules depending on the kind of services it offers. Every ring is arranged according to the criticality of its services, to its bandwidth, to the physical location of its servers and to their hardware type.

The servers are connected to one another through a VPN created by means of a software named tinc. All communications among the servers, from synchronization to mail routing, travel in an encrypted way through the VPN.

Layer two : the synchronization of our services

One of the basic aims of the R* Plan is to ensure that no services are suspended in case we are forced to put offline one or more nodes (e.g. because one node has been jeopardized).

To make this possible, we have had to sort out a mechanism for synchronizing all data and for easily redirecting all requests from the jeopardized server to a new one.

For the synchronization of services, the R* Plan makes use of different mechanisms:

Layer three : the users' data

As regards the users' data and the most substantial data sections connected to our services, it was not possible to rely on CFengine for their synchronization, due to the great amount of data it would have been necessary to transfer uselessly.

Data needing to be synchronized in more than one node (shared services data sections, some users' data, keys and certificates, etc.) are transferred via rsync. This applies for example to html pages distributed in several copies in the network, to backup services and to other stuff.

Every mailbox is physically located in one server, which has been chosen so as to balance the network load. At any time it is possible to move a a particular mailbox to a different server, by modifying an LDAP parametre. These movements will be totally transparent to the affected users. As with mailboxes, each website is located in one webserver of the network, and every website can be quickly moved, recovering all data from the backup copies available in the other network servers.

Layer four : users

One of the most important novelties of the R* Plan is the users' localization: actually, the R* Plan implies that all users are contained in one LDAP database (a database conceived of to be as efficient as possible when it is necessary to read many times and to write rarely).

The LDAP database contains all users' data, as well as the information relating to the different services linked to each user (the server containing her mailbox or site, her password, etc.).

Layer five : the services

In order to understand the implementation of the R* Plan from a technical point of view, you only need to grasp in which way the services are distributed among the different nodes.

Generally each service has been conceived of in order to be distributed among all nodes (usually by using round robin to sort the requests), while being independent from any particular node. Unfortunately, though, not all services allow the implementation of this pattern.

Let’s have a look at the main services:

Substantially, every single node of the network serves a share of websites and a share of mailboxes. If a server happens to be opened by someone, all configurations and the proportion of websites and mailboxes it used to host will be transferred to another node, thus avoiding a communication breakdown.

JavaScript license information