Eric Thune from Velocity Compute
In this podcast, Eric Thune from Velocity Compute describes how the company’s PeerCache software optimizes data flow for HPC I/O Scale Out as well as Hybrid Cloud Bursting.
PeerCache decouples compute and storage to deliver HPC I/O scale out with extreme file performance through peer-to-peer caching. The lines between file I/O and memory I/O are blurring – PeerCache uses NVMe today, but instead of building a distributed file system, it virtually projects entire enterprise filesystems onto a local flash device on the compute node. File extents are loaded dynamically on application demand, and ultra fast cache fill is provided using highly parallel peer to peer networking.
PeerCache delivers very high performance block and shared file services via instant, metadata updates to all peer nodes and highly parallel P2P content delivery between nodes as well as local writes. The file extent model allows PeerCache to bidirectionally transfer data and resultant deltas between on-prem and cloud, data center to data center or cloud to cloud with the lowest possible latency and minimal egress cost.
By using PeerCache to deliver hybrid cloud bursting, development teams can quickly extend their existing on-premise compute to burst into the cloud for elastic compute power. The power of the virtual projection guarantees that the on-premise workflows will run identically in the cloud, without the need for retooling, and the resultant data can be incrementally moved back to the origin as required.
PeerCache also eliminates data silos and duplicate data by keeping copies in the caches and only storing unique changes onto existing storage such as SAN, NAS, Lustre, Ceph, or Cloud Blockstores.
