CINES Chooses Bacula Enterprise Edition
Jérôme, please introduce us to CINES in a few words!
“The CINES center for higher education and research, has the following missions:
– intensive numerical computation,
– the long-term archiving of electronic data
– the hosting of nationwide computer platforms whose references are available on the CINES website (www.cines.fr).
What specific problems were CINES trying to solve?
“We needed to renew our data backup and recovery solution. Our current one worked well, in that it was stable and effective, but its overly-complicated licensing system continued to lead to endless “re-adjustments” of cost. We had to take into account many parameters such as the type of user, the processor, data volume, etc., without any way to check that we were in compliance. In short, we wanted an effective backup system that had a more simple licensing policy.”
How and why did you choose Bacula Enterprise Edition backup solution?
“As a result of the tender, we studied all the answers that came in. We grouped together the main backup software players. As with any tender, we have two evaluation criteria: the technical characteristics and the financial aspects. On both of these points the Bacula Enterprise Edition response presented by teïcée got the best rating”.
How would you describe the scope of your project?
“There are about 150 backed up workstations divided into two categories: workstations (mostly Windows) and servers (mostly Linux and CentOS). We save in “file” mode and the total saved volume is in the order of 150 TB, including retention. Bacula is installed on a physical server with a 10GbE interface for the data network and an SSD RAID for spooling data (a buffer before writing to tapes). The server pushes the data to two robotics that are shared with other applications. In each robotics three readers of type IBM 3592E06 and LTO4 and about 150 cartridges each type are dedicated to backup.”
How did the implementation of the Bacula solution take place?
“We had an installation service, product configuration, as well as training on how to use it. Subsequently, we benefited from substantial advice either by the Bacula Systems ticket management system or by directly contacting téïcée. We also participate, as part of a training plan, in Bacula’s Administrator I courses and then Administrator II.”
“The team came onsite site for deployment and worked on the physical installation of the machines, the operating system configuration, the Bacula software installation and configuration, and delivered the training “Administrator Level 1″. The installation operation lasted one week, after which the service was put in production.”
What is your assessment of this project?
“In technical terms, we see a real benefit from the fact that Bacula Enterprise is based on an Open Source solution: for example, there are many instances documented on the internet which describe how the Open Source angle makes it easily possible for us to do prototyping and experimenting at will. This has a positive impact on deployment time compared to a “black box” or “owner” solution.”
“In financial terms, some of the possible solutions we eventually chose not to use were unreasonably expensive. In fact, the cost of the next cheapest solution was two times higher than that of Bacula offered by teïcée. And the cost of the most expensive solution was 25 times higher than the cost of Bacula! Bacula System’s licensing is mainly related to the expected level of support and backed up machines, so it’s easy to make cost projections and forecasts. In terms of performance, the counterpart of the time invested in deeply understanding the software is that we are at present very confident in this solution and in our ability to administer it. The solution works and we can customize it to use it exactly the way we want.”
Using Bacula’s Flexibility to be Creative
“The HPC environment does not define what part of the data is important, so we therefore need to backup disks in their entirety – and without stopping production. To achieve this, we had to write scripts that are launched by the scheduler. Each “rack leader” is made of a pair of hypervisors that have a software RAID for the filesystem and LVM for the virtual machines. At the beginning of the job, the RAID is split and LVM snapshots are triggered. The structural information is then backed up on disk volumes and the system and lvm snapshots are backed up as block devices. At the end, the RAID disk is “re-added” to the RAID (this is very fast, especially when compared to a full, or hardware rebuild), and LVM snapshots are freed and merged. This way, when a “full backup” restore is required, the metadata are retrieved from disk volumes (fast), then the block devices are restored “in place”. “Without Bacula’s flexibility, this would had not been possible” said Jérôme.Jérôme’s conclusion: “Overall – the entire solution is especially reliable, scalable and powerful. It really meets our needs”.
Modules and Tools Adopted: