Have a nice day!
Fabio

1. Install some graphical tool for collection of bandwidth data in your server (like Cacti, Munin, Zabbix ecc.)
2. After some time (the longer the better, for statistical purposes) take the average bandwidth consumption, often given in kbps (Kilo Bytes per Second)
3. Convert kbps in GB per month and this where this post become useful :)

So to convert kbps in GB/Month we can simply obtain a constant from the following logic steps:

• 1kbps = 1KiloBitperSecond
• 1KiloBitperHour = 3600 kbps
• 1KiloBitperDay = 3600*24 kbps
• 1KiloBitperMonth = 3600*24*30 kbps
• 1Bit = 1Byte/8
• 1Kilo = 1Giga/1.000.000

So the constant will be 3600*24*30/8/1.000.000 = 0.324

For example 150kbps of bandwidth will generate 150*0.324=48.6GB per month traffic

or conversely, for example 80GB per Month will be 80/0.324 = 246.91 kbps on average throughtput

Therefore the constant is 0.324 and as a rule of thumb you can remember 0.3 or one/third ;)

Ubuntu 9.04 Server Edition makes it easy to experiment with cloud computing. Eucalyptus, an open source technology which is included in Ubuntu as a technology preview, enables you to use your own servers to deploy, experiment and test your own private cloud that matches the Amazon EC2 API. You can dynamically create virtual machines, configure multiple clusters into a single Cloud and even provide an EBS (elastic block storage) equivalent and an S3 compatible storage manager.

My needs are trying to implement Cloud Computing myself with some free software.

After searching around the web some days I decided to try the Rocks+Eucalyptus solution.

This document try to cover and log all the aspect that in the beginning where “not so obvious” or “not so clear” to me, saving time for documentation for me in the future and for you guys out there in the community ;)

Rocks is a “Clustering Suite” available at http://www.rocksclusters.org, today it is at version V, and we all know that when dealing with non commercial software being at version “5″ means a big and long effort with a lot of history and experience, in fact, if I’m not wrong, Rocks development started in the may 2000. Their goal was delivering the computational power of clusters to a wide range of scientific users, and they made many “Rolls” to do that, Rocks is the base and Rolls are the application running on it, for our purposes what we need is the Xen Roll.

Eucalyptus is a free Cloud Computing implementation available at http://eucalyptus.cs.ucsb.edu/

This test of the Eucalyptus Cloud Computing implementation on top of a Rocks V cluster has the following topology: 1 frontend and 2 computing nodes configured as VM container, the idea is that Rocks Clustering suite has a Xen roll and I wanted to try it out to understand its features and limits.

On Rocks you can have two types of nodes, Frontend and Compute, as simply understandable, frontend is the “director”, the “center of operations” and the machine that will delivery services, while compute nodes are the ones that will do the work, in our case hosting the Xen VMs.

The frontend has two NICs and has been installed with Rocks V frontend software, to do that, I downloaded the Rocks installation CD, full jumbo DVD distribution from here and simply started the boot process entering “frontend” at the boot prompt, the installation went really smooth, I “simply and dumbly” selected all the rolls in the DVD as this was a test drive and I didn’t want to loose time with dependency problem later, btw: on the frontend you need at least 2 GB RAM.

On the frontend node, the NICs are configured to be on two nets, the external production net, being on my LAN with 192.168.1.0/24 address and the internal network with 10.10.10.0/24 address.

For installing the computing or vm-container nodes simply follow the very well Rocks installation howto.

After launching the insert-ethers command, choose vm-container as node type, and repeat the installation for every physical node.

As soon as you reboot nodes you should be able to see them inside the director’s Ganglia tool, reachable connecting from localhost on the director Apache server with a local browser or enabling the network firewall rules for access to that from network.

Now that the Rocks cluster is on, it’s time to install the Eucalyptus software as described in the Eucalyptus Administrator Guide, I used Eucalyptus version 1.2.

First I made some test with Eucalyptus Public Cloud, than I installed my own Cloud on the Rocks cluster :)

If you use some old hardware on your nodes, remember to lower the minimum required RAM needed in your instances to work inside the nodes /etc/xen/xend-config.sxp, otherwise when you will query your cloud for availability you will get a 000/000 usable resources.

It works pretty well for me and Amazon’s ec2 tools are fine, unfortunately it just lacks native HA implementation.

Feel free to enrich this post with your personal experience and considerations.

Have a nice day ;)
Fabio