PernixData chap: We are to storage as Alfred Nobel was to dynamite

Snazzy memory tech demolishes DRAM volatility restrictions, says chief techie

Local hero

El Reg: But application data still has to be written to persistent storage. How does that work with DFTM?

Frank Denneman: This is true, but the key is that this operation is completely transparent to the application. Once the data is written to the local memory and the remote memory of another host in the cluster, the write operation is completed.

This architecture removes as many moving parts as possible. It writes to the local memory and simultaneously writes to remote memory across the Ethernet network to complete the synchronous replication.

Ethernet networks use a cross-bar architecture, which allows communication between nodes to operate without being impacted by communication between other hosts. This provides the fastest, most consistent method of communication between nodes and it scales very well.

In addition, DFTM uses a lightweight protocol for this data transfer: NFS and iSCSI are old protocols which weren’t designed for the speeds that are required by modern enterprise IT. Network bandwidth advancements are stellar at the moment, 40GbitE is just around the corner and 100GbitE will follow soon after.

Writing close to memory speed and having that bandwidth available across nodes in the cluster allows you to create storage speeds that are not achievable with traditional storage architectures.

El Reg: Thank you Frank.

To understand more about his thinking on memory and application acceleration read his blog. ®