Guaranteeing no lost frames is a matter of having sustainable bandwidth. We guarantee that with these means…
- Metadata access path is separated from data access path (don’t want eating sugar cubes having to wait for eating elephants)
- Data access path is parallel – able to go directly to separate servers (want to be able to eat elephants concurrently).
- Data access path can use multiple TCP/IP connections to aggregate bandwidth (overcomes TCP/IP limitations)
- Data access path can use multiple network paths/ports using bonding or #3 above (to different IP addresses).
- Data access path can use RDMA
- Data access path can bypass networking altogether if the media is on the same physical server as the application using it (we call this DA-DS for “direct access data store”, or sometimes “protocol bypass”). This is especially interesting for hyper-converged scale-out infrastructure, such as a Kubernetes cluster.
- When NFSv3 or SMB is required,
- The DSX-Portal is a stateless, scale-out service, allowing scalable performance.
- Tha DSX-Portal benefits from items 2-5 above. #5 is particularly interesting, since it is very likely that the portal has local storage. The access to that is, thus, direct.
The bottom line is that with the above ^ we can
- Maximize the output per server node/cloud instance – limited only by what the HDDs/SSDs and networking is capable of
- Get linearly scalable performance with ever addition node added. So, if one node can get 8GBytes/sec (which, BTW, some can), then 10 nodes can get 80GBytes/sec.