Storage Controllers – Your Questions Answered

The term controller is used constantly, but often has very different meanings. When you have a controller that manages hardware, there are very different requirements than a controller that manages an entire system-wide control plane. You can even have controllers managing other controllers. It can all get pretty confusing very quickly. That’s why the SNIA Ethernet Storage Forum (ESF) hosted our 9th “Too Proud to Ask” webcast. This time it was “Everything You Wanted to Know about Storage but were Too Proud to Ask: Part Aqua – Storage Controllers.” Our experts from Microsemi, Cavium, Mellanox and Cisco did a great job explaining the differences between the many types of controllers, but of course there were still questions. Here are answers to all that we received during the live event which you can now view on-demand.

Q.Is there a standard for things such as NVMe over TCP/IP?

A. NVMe™ is in the process of standardizing a TCP transport. It will be called NVMe over TCP (NVMe™/TCP) and the technical proposal should be completed and public later in 2018.

Q. What are the length limits on NVMe over fibre?

A. There are no length limits. Multiple Fibre Channel frames can be combined to create any length transfer needed. The Fibre Channel Industry Association has a very good presentation on Long-Distance Fibre Channel, which you can view here.

Q. What does the term “Fabrics” mean in the storage context?

A. Fabrics typically applies to the switch or switches interconnecting the hosts and storage devices. Specifically, a storage “fabric” maintains some knowledge about itself and the devices that are connected to it, but some people use it to mean any networked devices that provide storage. In this context, “Fabrics” is also shorthand for “NVMe over Fabrics,” which refers to the ability to run the NVMe protocol over an agnostic networking transport, such as RDMA-based Ethernet, Fibre Channel, and InfiniBand (TCP/IP coming soon).

Q. How does DMA result in lower power consumption?

A. DMA is typically done using a harder DMA engine on the controller. This offloads the transfer from the host CPU which is typically higher power than the logic of the DMA engine.

Q. How does the latency of NVMe over Fibre compare to NVMe over PCIe?

A. The overall goal of having NVMe transported over any fabric is not to exceed 20us of latency above and beyond a PCIe-based NVMe solution. Having said that, there are many aspects of networked storage that can affect latency, including number of hops, topology size, oversubscription ratios, and cut-through/store-and-forward switching. Individual latency metrics are published by specific vendors. We recommend you contact your favorite Fibre Channel vendor for their numbers.

Q. Which of these technologies will grow and prevail over the next 5-10 years…

A. That is the $64,000 question, isn’t it? J The basic premise of this presentation was to help illuminate what controllers are, and the different types that exist within a storage environment. No matter what specific flavor becomes the most popular, these basic tenets will remain in effect for the foreseeable future.

Q. I am new to Storage matters, but I have been an IT tech for almost 10 years. Can you explain Block vs. File IO?

A. We’re glad you asked! We highly recommend you take a look at another one of our webinars, Block vs. File vs. Object Storage, which covers that very subject!

If you have an idea for another topic you’re “Too Proud to Ask” about, let us know by commenting in this blog.

Storage Controllers – Are You Too Proud to Ask?

Are you a control freak? Have you ever wondered what the difference was between a storage controller, a RAID controller, a PCIe Controller, or a metadata controller? What about an NVMe controller? Aren’t they all the same thing?

On May 15, 2018, the SNIA Ethernet Storage Forum will tackle these questions and more in “Everything You Wanted To Know About Storage But Were Too Proud To Ask – Part Aqua: Storage Controllers.”  In this live webcast, our experts will take an unusual step of focusing on a term that is used constantly, but often has different meanings. When you have a controller that manages hardware, there are very different requirements than a controller that manages an entire system-wide control plane. From the outside looking in, it may be easy to get confused. You can even have controllers managing other controllers!

In Part Aqua we’ll be revisiting some of the pieces we talked about in Part Chartreuse, where we covered the basics, but with a bit more focus on the variety we have to play with:

  • What do we mean when we say “controller?”
  • How are the systems managed differently?
  • How are controllers used in various storage entities: drives, SSDs, storage networks, software-defined
  • How do controller systems work, and what are the trade-offs?
  • How do storage controllers protect against Spectre and Meltdown?

I hope you will register today and join us on May 15th to learn more about the workhorse behind your favorite storage systems.