The French deposits and consignments fund la Caisse des Dépôts, a state-run investment organisation, has opted for NetApp’s StorageGRID object storage to hold backup copies files from its NAS production arrays.
Object storage might appear a surprising choice, given that it is most suited to long-term and archive storage or for files for use by web applications. Here, in fact, the challenge is to be able to restore any file rapidly in case of an outage or disaster.
“The principal goal of the project was to deploy the most eco-friendly backup solution possible,” says Allan Denis, storage and systems engineer with the la Caisse organisation’s IT subsidiary, the GIE Informatique CDC. “For that reason, object storage triumphed over other storage technologies. But the difficulty was to find a solution that interfaces with production NAS arrays that use the NFS protocol.”
Another difficulty was to resolve the question that the NAS used by la Caisse is distributed over several datacentres for high availability on big data workloads with sensitive data. It was therefore necessary to take into account the existence of some delicate parameters and exercise some gymnastics at the network level.
NetApp continuity between storage protocols
The GIE Informatique CDC is a longstanding NetApp customer. In 2014, it replaced its NetApp block access arrays with NetApp filers and was very satisfied with how simple the transition between the two systems was.
Before studying the market more widely, Denis looked into the possibility of being able to work with object storage from the systems already deployed. It turned out that NetApp had built this possibility into its Ontap operating system. Functionality called FabricPool allows StorageGRID object storage arrays to connect with existing storage as a simple additional storage tier.
“The advantage of this solution is that there isn’t much you need to do to make it work,” says Denis. “We only needed to create rules at the level of Ontap to define at what point NAS data is pushed to StorageGRID. This rule sets up a job that executes at the required time to convert files on-the-fly to object mode.”
Another advantage with the NetApp object storage product is that it uses an algorithm, StorageGRID Webscale, to share data between different sites that are geographically distant and connect via the WAN. This functionality exactly fitted the requirements of the GIE, however.
“We use private fibre between all our sites and this works very well with the NAS cluster,” says Denis. “A master NAS can share its data with a master StorageGrid Webscale, to the right places and according to the rules.”
A simple solution that performs
The system was five months in testing, installation and putting into production. The GIE opted for twin redundant StorageGrid arrays at each site, with each in a 2U, 12-drive form factor.
For security reasons, Denis did not disclose the number of sites equipped like that, but total capacity comes to 380TB.
“We are impressed with the efficiency of the system,” says Denis. “StorageGRID arrays have spinning-disk hard drives, but according to our measurements, access to data is practically as fast as to our NAS arrays that have flash drives.”
Denis points out that, from the network point of view, traffic passes via conventional F5 load balancers. “StorageGRID takes charge of all the logic of distribution and it hasn’t been necessary to invest in dedicated S3 gateways,” he says.
Another point of satisfaction concerns maintenance. Updates are carried out without interruption to service and to each controller in turn. “This was already the case with our NAS and it’s something we really like from NetApp,” says Denis.
Each StorageGRID controller can support up to 2.14PB of capacity. The GIE intends to use its object storage for other workloads. As well as serving as storage for backups for big data, the system could be used for cold data from other NAS storage which is no longer needed in production.
“Other use cases don’t require investment in another solution,” says Denis. “We just need to create new object storage buckets in StorageGrid and to append new rules.”