The server-side components of the CASMACAT workbench have to be installed on a Linux machine. This machine then can be accessed over the web with any Internet Browser.
Get a Machine
A modern PC is all you need.
Please consider the following regarding the machine's configuration:
- Disk space: Commonly available 1 TB hard drives should be sufficient for most needs.
- CPU: Since components of the workbench can run multi-threaded, having multiple cores can be beneficial. It is only really a requirement if you expect to have multiple people accessing your CASMACAT server at the same time.
- RAM: If you are running a machine translation server on the machine, then the more the better. You will need at least 4 GB, better 16 GB or more.
The most computationally demanding process will be training a machine translation system on large amounts of data. If you, say, build a system on your own data of a few million words, you should be fine with regular hardware. However, commonly used large scale systems using 100s of millions of words of parallel data and billions of words of language model data are typically trained on machines with dozen(s) of cores and over 100GB of RAM. You may train a system on a different machine, so your run-time machine will not need to be as powerful.
Choose a Linux distribution
There are several Linux distributions available, and all should work. They place files in different locations, so some of the instructions below need to be adjusted. This installation guide was developed on Ubuntu.
If you want to use your existing Windows desktop machine as a CASMACAT server, then you can install Linux on a separate partition. This means, however, that you cannot access Windows on the machine at the same time.
Another option is to install Linux in a virtual machine within your Windows environment. We have not tested this.
The CASMACAT workbench uses a standard set of tools: the Apache web server, the programming language PHP, and the mySQL database. All these tools are part of a standard Linux distribution but may need to be installed on demand.