16 and 28 THREADS per cpu, 8 and 14 cores respectively.
Curious why they go such low core counts when 18 and 22 core xeons have been out a while already.
Datrium has announced a rack scale product to ease large-scale deployment and management software providing virtual machine-focussed protection, replication, copy data management and archiving with a roadmap to public cloud support. The company provides DVX storage arrays in which the controller functionality runs in accessing …
Hi Nate,
The Datrium DVX supports 3rd party servers with flash as compute nodes by uploading our software to run on ESXi. That platform approach was how we approached the market over last year, and we've gotten very good at it. It's "Open Converged."
We're now providing turnkey compute nodes in the middle of the x86 bell curve of price/performance, but we're not trying to offer as many server configurations as a hardware vendor because it's Open Converged. This doesn't just go for core counts. If you need quad-socket servers or blade servers, most hyperconverged vendors leave you to your SAN. DVX can fully integrate them because its not a silo proposition.
We'll grow our turnkey configurations over time based on demand.
Thanks for the nice write-up, Chris. My quick compare of Datrium DVX to HCI can be found here: goo.gl/PRsLVn
DVX doesn't have an array inside. Arrays are, as Gartner would say, 'external controller based storage' -- array RAID is done across a fabric from hosts in a shared controller that connects to drives directly. This makes the controller a bottleneck and a change-management obstacle.
DVX runs erasure coding and other data services on each compute node, so it scales naturally with workload. It writes to a separate data node to enable automatic performance isolation and separate availability domains, which simplify management at rack scale and allow much more predictable low-latency response. Additional notes on the benefits of this architecture for NVMe are here: goo.gl/kpyFIj