It saved 42% in energy costs, 50% in datacentre rackspace and got 4.5PB where previously it had 3.5PB in its NAS storage hardware refresh.
That’s the result achieved by French TV giant TF1, which replaced 80 nodes of Dell Isilon scale-out NAS storage with 40 new PowerScale nodes from the same array maker.
The upgrade comes as TF1 lays foundations for TF1+, which aims to rival the likes of Netflix as a content platform and needed new capacity to store 15,000 hours of media.
“And it’ll be higher definition,” says director of IT operations for TF1 Jean Gras. At the same time, he adds: “We need to take into account new social responsibilities, especially the aim to significantly reduce carbon footprint by 2030.”
Scale-out NAS is file access storage, and well-suited to audio-visual content that’s made up of quite large files to which often sequential access is required. Scale-out NAS allows for the creation of a very large file system that runs across multiple hardware nodes, with adds and changes to capacity on the fly. As a key method of unstructured data storage it is rivalled by object storage.
TF1 has set itself some ambitious challenges. The company – which has 3,000 staff and is a producer and broadcaster of audio-visual content – will offer the same movies and series as its competitors, but aims to set itself apart with news content. Key to free access on TF1+ is that users will be invited to watch advertising segments when accessing content. There will also be a recommendation engine similar to that used by Netflix.
“Our content is distributed under a variety of formats from servers in AWS datacentres,” says Gras. “We looked at the possibility of storing all our original video files in AWS. But, according to our simulations, the cost of data storage that’s not really ‘cold’ in the cloud doesn’t compare with an on-site array.”
Reflection on storage offerings to meet the new challenges started in the second half of 2022.
PowerScale and Isilon are both made by Dell, with the former the evolution of the latter. Both are scale-out NAS that run the OneFS parallel file system to which nodes can be added to augment capacity, almost to infinity and under one namespace.
But TF1 did ask the question of whether to change storage hardware supplier.
Stick with OneFS to avoid costly skillset training
“Several elements caused us to stay with Dell,” says Gras. “The first was the availability achievable on OneFS. These arrays need to store rushes sent by journalists on the ground and make them available in just a few minutes for broadcast.”
“The second element is that it was evident that to stay on OneFS allowed us to profit from existing management and operational routines. We worked out that to change platform would have necessitated training between 15 and 20 people.
“The final advantage of staying with OneFS is that it would guarantee an uninterrupted migration,” he says.
As it happened, the new PowerScale nodes were easily added to the existing cluster, which reached capacity of 8PB for a while. OneFS was reconfigured so that any new writes went to the new nodes, and that the contents of the Isilon nodes went to PowerScale as and when that was possible and bandwidth was available.
The migration occurred non-disruptively thanks to the use of load balancing upstream of the nodes.
Migration started at the beginning of 2023, and lasted until spring that year. When it had completed, the old Isilon nodes were recycled by Dell, which suited TF1’s ecological aims.
Access times 10x more rapid
Of the 40 nodes deployed, 14 are of the model A3000L and 26 of F600. The A3000L runs SATA hard drives that can store large quantities of cold data at a better cost point.
The F600 arrays run flash drives to accelerate access. Data hits the F600s first before OneFS takes decisions about whether to stage it to A3000L capacity.
“We’ve noticed acceleration of 10x in speed of access,” says Gras. “The optimisation is particularly effective with the SSDs.”
Hosts access the storage cluster via classic Ethernet infrastructure. Storage nodes use Infiniband as an interconnect.
At the same time as the migration took place, OneFS also transferred all contents to a second disaster recovery (DR) site.
“Right now, we have to put in place a full DR plan,” says Gras. “It’s essential so that we can deal with a breakdown at the production site but also in case of cyber attack that we can restart activity with clean copies.”
“Dell is a very efficient partner when it comes to elaborating scenarios with us, and that includes understanding our data that resides with AWS.”