3 ms·
Something I don't understand: Why is it important that storage/compute be in the same box? Isn't it desirable for there to be a "service boundary" between bloc
by ahepp 4y ago
Something I don't understand:
Why is it important that storage/compute be in the same box? Isn't it desirable for there to be a "service boundary" between block storage, file storage, object storage, etc, and the consumers?
If that boundary is desirable, why would it matter where the storage resources are located? Wouldn't that be an implementation detail?
- tcbyrd 4y agoHyperconverged is all about software defined storage and compute. You can create those service boundaries all on one cluster and pool like nodes together to create one big mesh of compute and storage. The precursor were things like EMC and NetApp storage clusters which typically had 2-4 "compute nodes" with a rack full of direct attached storage. This created a massive problem come upgrade time, and the term was called "forklift upgrade" [1] for a reason. With HCI you can add and replace single boxes as needed. Also, latency is not an implementation detail for a lot of data-intensive workloads. We're talking differences of 1-10ms latency for Network Storage and 250-500µs for a local NVME SSD. [1]: https://www.architecting.it/blog/what-is-a-forklift-upgrade/ https://www.architecting.it/blog/what-is-a-forklift-upgrade/
- ElectricalUnion 4y ago* dense (in rack U's) computing/storage; * potentially compute and storage are colocated on the same chassis, providing maximum PCIe bandwidth and lower latency; * reduced need to buy and mantain dedicated/offband storage networking solutions; * You buy less half-filled, special purpose boxes.