Tuesday, May 26, 2020

Pavilion introduces S3 storage

Pavilion Data Systems, emerging leader in high-end storage, shakes the market with the introduction of S3 storage.

I had the privilege to attend a webcast this morning pre-announcing S3 as the third access method exposed by its Hyperparallel storage array. And honestly it was a surprise.

The design of the product allows up to 20 controllers - 10 cards of 2 controllers - each controller can expose one type of protocol among block with NVMe over Fabrics (over TCP or RDMA on Ethernet - RoCE - or IB), file here NFS (v3 or v4) or object with S3.

For S3 it means one S3 namespace per controller and a namespace can't span other controllers in phase 1. In other words, namespaces are not cross controllers yet.

Phase 2, later this year, will allow namespaces to span controllers but it will be only one namespace per controller.

Phase 3, supporting all capabilities introduced in phase 1 and 2, multi-chassis namespace will be available.

But similarly to one controller exposing one protocol, a controller can manage only one namespace. The capabilities of multiple namespaces is achieved via the consideration of multiple controllers and therefore you can assign different namespace to different tenant.

Also a file written via NFS can't be read via S3 and vice-versa as you need to copy data across namespaces and protocols, it consumes extra capacity - double here and even triple when SMB will be available - and potentially it creates a data divergence. Pure Storage with FlashBlade behaves the same but Qumulo and VAST Data offer this multi-protocols access.

Pavilion strategic direction seems to be around unified storage now delivering top performance numbers for multi-protocols instead of only a top block array. I understand that the idea is to make available the array and its enormous bandwidth to any protocols and workloads. Good idea, market traction will tell us how the product lands in the field and if this strategy resonates...


0 commentaires: