QLogic: Software FCoE can't cut it
El Reg: Will it be feasible to mix storage networking traffic with general Ethernet traffic?
Henrik Hansen: Without enterprise management applications, CPU-offload and a proven storage driver stack it would be a stretch to believe this could be successful.
El Reg: Will physical Fibre Channel users move to software FCoE initiation and bypass expensive CNAs altogether?
Henrik Hansen: No – enterprise Fibre Channel users will want to retain the reliability that only their existing proven storage stack can offer. Price differentiation may not be as much as you believe, and what you are paying for is piece of mind that your storage infrastructure will work as advertised.
In comparison, with an Intel solution, VM density, and application workload will be sacrificed, which comes with additional expense. The purchase of more CPU or less VMs per physical server will drive up data centre costs above the difference of a NIC compared to a CNA.
El Reg: What effect will 16Gbit/s Fibre Channel (FC) have on this, being faster than 10Gbits Ethernet?
Henrik Hansen: Not all data centres are interested in convergence. Some believe that storage traffic is best separated and managed by storage administrators. These data centres will remain on FC and migrate to 16Gb FC in the future.
El Reg: Has Intel miscalculated what the FC market needs, or has QLogic miscalculated?
Henrik Hansen: The 10Gb Ethernet market is a highly competitive market with many new players as well as large, proven players. QLogic believes that the value of storage over Ethernet is held within the storage driver stack. Many others believe this philosophy is true, including Broadcom who attempted a hostile takeover of Emulex for its driver stack. OEMs seem to hold this same belief as CNAs have won all storage OEM design wins to date.
El Reg: Meaning Intel has miscalculated, at least as as far as the enterprise data centre is concerned, and possibly elsewhere too. What effect will the current situation of FCoE standards have on Intel's SW FCOE chances?
Henrik Hansen: The standards will not drive adoption; O/S certifications, storage qualifications and Hypervisor adoption will, and FCoE software initiators are far behind CNAs with their proven driver stack and support by all fore-mentioned criteria.
SW FCoE, if it is taken up, will be taken up by the SMBs, like iSCSI, and not by enterprise data centres. The latter will prefer HW CNAs because they will be certified by OEMS, work with server O/S and hypervisor system software, support OEM partitioning, provide better and easier management and help to keep storage traffic hygienically separated from general LAN traffic.
It may be a bit of a stretch to assert that FCoE standards will not drive adoption. After all, the lack of such standards has been cited as a factor slowing FCoE adoption.
The QLogic takeaway here is that, in the enterprise data centre, Intel's SW FCoE stack won't cut it. The Reg reckons that outside and in SMB land there is existing and proven iSCSI. What need is there therefore, for SW FCoE? What does it offer that iSCSCI does not for the SMB market? ®