I run lots of Linux servers. I create them, install some stuff, mess around with them, forget them, come back to them…and forget my credentials. My life used to look like this:

$ ssh -i ~/steveko.pem steveb@115.146.83.268
Permission denied (publickey).
$ ssh -i ~/steveko.pem sbennett@115.146.83.268
Permission denied (publickey).
$ ssh -i ~/stevebennett.pem steveb@115.146.83.268
Permission denied (publickey).
$ ssh -i ~/steveko.pem ubuntu@115.146.83.268
Welcome to Ubuntu 12.10 (GNU/Linux 3.5.0-26-generic x86_64)
...

This got really tedious. You can’t memorise many IP addresses, so you’re constantly referring to emails, post-its or even SMSes. Then you rebuild your server, the IP address changes, and you’re lost again.

And because some of the servers are administered by other people, I can’t always choose my own user name, so more faffing around.

So, here’s my solution:

A naming convention for servers

Give each server a name. Ignore the actual hostname of the server, or what everyone else calls it. My convention goes like this:

<infrastructure>–<project>[–<purpose>]

Examples:

• nectar-tugg-dev: A development server for the TUGG project, running on NeCTAR Research Cloud infrastructure.
• rmit-microtardis-prod: A development server for MicroTardis, running on RMIT infrastructure.
• nectar-tunnelator: A side project “tunnelator” running on NeCTAR Research Cloud infrastructure. Small projects only have one server.

The key here is minimising what you need to remember. If I’m doing some work on a project, I’ll always know the project name and whether I want to work on the prod or dev server. Indeed, it’s an advantage to have to specifically type “-prod” when working on a production machine.

Put all IP addresses in /etc/hosts.

When I create a server, or someone tells me an IP, I immediately give it a name, and store it in /etc/hosts. The file looks like this:

##
# Host Database
##
127.0.0.1 localhost
255.255.255.255 broadcasthost
::1 localhost 
fe80::1%lo0 localhost
115.146.46.269 nectar-tunnelator
136.186.3.299 swin-bpsyc-dev
...

This has the huge advantage that you can also put those names in the browser address bar: http://nectar-tunnelator

If a server moves location, just update the entry in /etc/hosts, and forget about it again.

Put all access information in ~/.ssh/config

The SSH configuration file can radically simplify your life. You have one entry per server, like this:

Host latrobe-vesiclepedia-dev
 User steveb
 IdentityFile /Users/stevebennett/Dropbox/VeRSI/NeCTAR/nectar.pem
 Port 9022

Notice how we don’t need to spell out the IP address again. And by storing the access details here, we can connect like this:

$ ssh latrobe-vesiclepedia-dev

So much less to remember. And because it’s so easy to connect, suddenly tools like SCP, and SSH tunnelling become much more attractive.

$ scp myfile.txt latrobe-vesiclepedia-dev
$ ssh latrobe-vesiclepedia-dev sudo cp myfile.txt /var/www

In reality, it gets even simpler. Most of my NeCTAR boxes are Ubuntu, with a login name of “ubuntu”. The “nectar-” naming convention proves valuable:

Host nectar-*
 IdentityFile /Users/stevebennett/Dropbox/VeRSI/NeCTAR/nectar.pem
 User ubuntu

That means that any NeCTAR box using that key and username doesn’t even need its own entry in .ssh/config:

$ ssh nectar-someserver