Well, FC-NVMe. Did this lightning-fast protocol just get faster?
New wine in old bottles; NVMe fabrics can use Fibre Channel instead of replacing it
+Comment A walk through the vast and spacious exhibition arena at HPE Discover in London can bring you to Cavium's stand. There Roberto Angelo Polacsek, a senior account exec, will tell you why he believes NVMe over Fibre Channel will be important. Cavium bought Fibre Channel HBA vendor QLogic recently, and Polacsek says its 16Gbps HBAs can be upgraded to add NVMe fabric support, FC-NVMe, for free.
Why is that important? A QLogic white paper (PDF) says: "Fibre Channel Protocol (FCP) is a transport protocol that predominantly transports SCSI commands over Fibre Channel networks. With the advent of NVMe, FC is transforming to natively transport NVMe and this new technology is called FC-NVMe."
These SCSI commands assume that the target media is disk and uses the Advanced Host Controller Interface (AHCI) protocol/command set. So Fibre Channel SANs transport SCSI/AHCI requests, encapsulated inside frames, across the basic Fibre Channel fabric, and these requests and their translation waste time when the target media is flash. By having an end-to-end NVMe fabric link between accessing servers and the SAN, the need for SCSI/AHCI goes away and the data access latency drops.
An NVMe fabric link across Fibre Channel can coexist with existing SCSI/AHCI links across Fibre Channel, meaning that current SAN users can upgrade their SAN access to FC-NVMe at their own pace, on a LUN-by-LUN basis, and with no need to replace the existing physical Fibre Channel infrastructure with either Ethernet or InfiniBand. Broadcom/Brocade (PDF) is on board as well.
Fibre Channel SAN NVMe fabric adopters will need to be aware that the existing balance between Fibre Channel (SCSI/AHCI) IO load and SAN controller capability may well be upset and plan accordingly.
For an existing all-flash array user, whose array is connected by Fibre Channel to accessing servers, the prospect of software upgrading the link to NVMe must be appealing. It's like having a software upgrade to ludicrous mode in your Tesla.
Say the array is an all-flash 3PAR or Pure Storage product with a promised upgrade path to NVMe drives. When that upgrade happens you have the link upgrade to FC-NVMe sorted out in principle, and can turn your attention to the software stack in accessing servers to ensure that it can use NVMe over fabrics effectively.
HPE likes the idea of a composable infrastructure; it's what lies behind its Synergy product and the massive Machine project. Physical SANs were one of the first examples of this, decomposing storage from servers, and enabling it to be managed, operated, provisioned and consumed much more effectively than by having it implemented as captive storage inside servers. Hyper-converged infrastructure is putting that clock back.
We could say physical SAN survival in the flash era depends upon NVMe over fabrics. At NVMeF speed every LUN in the SAN is effectively local to every accessing server and it may be simpler to manage all your storage in one place instead of having it spread across hundreds of hyper-converged infrastructure server nodes. ®