Experts from the FCIA recently took a deep dive into the FC-NVMe protocol at our live webcast, “NVMe over FC: Deep Dive in Protocol, Architecture and Use Cases.” It was a fascinating technical discussion that attendees gave a 5-star rating. If you missed it, you can watch it on-demand on the FCIA YouTube Channel or on BrightTALK. The presentation slides are also available on the FCIA website.
We ran out of time to answer all the questions from the audience during the live session. As promised, here are our answers from our expert presenters.
Q. Why peer zoning / targeted zoning only is supported with NVMe over FC? Also, how smooth to migrate from classical SCSI FC to NVMe over FC, can it be done online?
A. This is implementation dependent. The way SCSI target LUN is presented vs. the NVMe namespace will differ depending upon the approach. At the end of the day, it’s a worldwide name, what it maps to is implementation dependent. And yes, migration from SCSI FC to FC-NVMe can be done online.
Q. I heard queues and performance directly depends on the amount of CPU cores on the host. Is that true?
A. The default number of the I/O queues will match the number of the cores of its’ CPU. But users can change that value to any number irrespective of the number of the cores. But, a higher number of cores will yield better performance.
Q. Why has NVMe over FC not been more widely deployed?
A. In almost all products that offer NVMe/FC, the traditional FCP version is still available. Users that are accustomed to using the traditional FCP connectivity methods can continue to use it, knowing that NVMe/FC is a new option. In the future, it is expected that users will become more accustomed to using the NVMe/FC feature and will develop that preference or will be forced in that direction as vendors begin offering only NVMe/FC.
Q. With existing FC environments, it appears the FC-NVMe is being slow to adopt by end-users in production even though it has been supported for some time now. What factors do you see as to why it is not being adopted more rapidly? What factors will drive a wider adoption of FC-NVMe and how long do you think it will take to replace SCSI in general?
A. Please see the answer to the question above. In addition, over time vendors could decide to add further integrations between the fabric and back end native NVMe storage devices, which could drive a further preference for the NVMe/FC feature advantage.
Q. With the explosion of AI, FC seems to have all the characteristics required for network AI needs. Would there be a use case for NVMe-oF with AI?
A. We have shown that NVMe/FC offers performance advantages over FCP for AI/ML workloads. Please watch this FCIA webcast on this topic for more information. The use case would be for Fibre Channel connected AI servers that require highly maximum performance for highly threaded transactional workloads.
Q. What is the current market adoption percentage of NVMe/FC in the market today compared to FC-SCSI? How does market adoption for NVMe/FC compare to NVMe/TCP?
A. Specific usage of NVMe/FC is not tracked by market analysts, but we can make an estimate of the HBAs and switches that are NVMe/FC capable that are likely in-service today. According to the 2023 Quillin Research SAN Forecast, 48% of FC ports shipped in the last five years are either Gen 6 or 7 hardware levels, meaning they are NVMe/FC capable. The other adoption factor is to look at which storage arrays offer NVMe over Fabrics. NVMe/FC was the first NVMe-oF transport to become almost universally available on most enterprise class all-flash storage arrays, meaning that these arrays can plug into any Gen 6 or 7 SAN and utilize the FCP or NVMe/FC feature today. By contrast, fewer storage arrays offer Ethernet as an NVMe-oF block storage connection, those that do may have standardized on either NVMe over RoCE or NVMe over TCP but usually not both. The conclusion here is that NVMe/FC today is far further ahead in market availability as a fully tested, supported and interoperable solution compared to NVMe/TCP.
Q. What are some of the inherent security advantages to NVMe/FC versus Ethernet-based protocols?
A. Fibre Channel fabrics are inherently more secure due to the “Air-Gap” nature of not sharing the Ethernet physical fabric.
Q. Is the SLER advantage only applicable to NVMe-FC or does this work in the Ethernet/TCP-based implementations of NVMe?
A. SLER exists for SCSI FCP and NVMe/FC. Any re-transmission is done using the TCP protocol.
Q. Is there IBM AIX support for NVMe/FC yet?
A. AIX HBA drivers are not supplied by the HBA vendors but by IBM. We advise that you ask your IBM representative. An internet search reveals information that seems to confirm that NVMe/FC is now supported on IBM Power and AIX.
Q. What is the minimum FC-NVMe speed supported or qualified?
A. FC-NVMe is not tied to a speed. In general, vendors started support for FC-NVMe on Gen 6 hardware (32GFC), so typically anything that supports 32GFC or newer will support NVMe/FC, consult your OS and equipment vendors for specific details on support.
Q. Do you have any examples of customers using NVMe/FC for Splunk hot/warm data, and if that gained a performance improvement?
A. We do not have examples to share. We suggest you check with your Splunk contact.
Q. How do you see the competition from Ethernet?
A. Since the inception of the FC transport protocol in 1994, it has been the gold standard for enterprise storage applications. 30 years of innovations like congestion control, analytics, and automation have made FC rock solid. FC has major market share for mission-critical data center applications, and we expect that to continue into the future.
We encourage you to keep in touch with the FCIA. Make sure you don’t miss any future educational webcasts by following us on LinkedIn and @FCIAnews.
