Massive Media dumps Amplidata, makes eyes at Caringo
Amplidata exec hits back: 'Nonono, you've got it all wrong'
Object storage supplier Amplidata has been replaced by fellow supplier Caringo at Massive Media, which bills itself as the world's largest social discovery site.
Massive runs two social networking websites: Twoo, which was launched in 2011, and Netlog, which appears to share the same basic concept as Facebook. Between the two sites, Massive claims to have over 110 million users.
Massive started out using its own self-designed system to store the millions of photos and files from its users, before switching to MogileFS. Yet neither system, a Caringo press release says, "provided the scalability the company needed”.
Moving to Amplidata’s AmpliStor product didn't appear to work out for Massive as its sites grew: "The efforts of scaling out were getting slightly bigger than the gains."
Massive’s system has to cope with hundreds of concurrent reads, writes and deletes per second. Software updates must be carried out with no overall system downtime and the recovery time of a single node had to be less than a day. Massive also needed to provide linear storage capacity scalability and near-linear scalability of throughput capacity.
Nicolas Van Eenaeme, Massive Media's director of technology, said: "At first, everything was going OK with our previous storage, but as we grew, it started to have its hiccups. Also, there was a huge storage overhead in the previous systems we used and all of them had in one way or another a single point of failure."
He switched the company to Caringo's CAStor and says he likes its shared-nothing architecture, performance and scalability.
"If you need more throughput or storage, you just add more nodes,” Eenaeme said.
The company is now migrating all of its legacy storage to Caringo.
It stores each object multiple times on three clusters:
- Cluster 1 has 60 nodes on 30 physical x86 servers, and 7,200rpm SATA drives
- Cluster 2 has 32 nodes with solid state drives
- Cluster 3 has 42 nodes, with 7,200rpm SATA drives again
Cluster nodes are interconnected with 1GbitE connections.
The grand storage total is 78TB of user pictures and thumbnails, and you can get a Storage Strategies account of the new system here (registration required). According to this account, "the capacity requirements of Massive Media's network are expected to double every two years."
Take that, Amplidata … or not. Their CTO and co-founder Wim De Wispelaere says: "I talked to Nicolas … He feels he is quoted incorrectly," and goes on to explain:
AmpliStor, with its PetaByte scalability for Big Data, has been deployed at Massive Media to store hundreds of TB of pictures and movies from users. Over time, and with the launch of new services such as Twoo, smaller files and thumbnails became the predominant element being stored.
In the article Nicolas states hundreds of million files in 29 TB. Nicolas measured that erasure coding, while providing superior data protection, didn't give him the storage savings for these small files and thumbnails. He decided that an object storage system that is storing full copies of those files and thumbnails on a small number of disks better suits this new use case.
Amplidata has since added small file optimisation by introducing a new small file policy in its AmpliStor 3.1 release March 2013. This small file policy enables up to 20 times more parallel small file retrieval operations per second. Amplidata [3.1] now efficiently supports small files and large files in a single AmpliStor system.
Dell's OEM use of Caringo's product recently ended, with the relationship reverting to a reseller one. ®