AMAX recently sat down with a team from Emerson Network Power to discuss Redfish, a server management spec intended to supersede IPMI. The goal of Redfish is to provide a management standardization across various server/storage platforms, networking and operating systems that can scale better than IPMI and give companies more flexibility and control of how they integrate and manage their data centers.
Note: Emerson (as referenced in the following in interview) has since been acquired by Vertiv.
AMAX: So the first question is a basic one. What is server management and why is it important?
Emerson Agustin: I’ll go ahead and take that. Server management is the command and control of your server infrastructure, usually through an out-of-band interface that’s separated from your production network. It’s important, because it provides a way to:
- Configure and control your systems remotely, and
- Manage critical alerts.
So if you have a problem, you can leverage that out-of-band interface to go in and figure out what’s going on and immediately take corrective action.
AMAX: What are major issues with server management in today’s data centers?
Emerson Agustin: One of the biggest problems today is the inconsistency of APIs and interfaces. Using different toolsets to manage your infrastructure is cumbersome. Operators want to use one set of tools across their data center, and porting toolset to match proprietary interfaces is a pain.
AMAX: So the current standardization is IPMI. In your opinion, why is it no longer a good fit for today’s data center landscape?
Emerson Agustin: IPMI was actually the first attempt at providing a common interface for data center infrastructure management, particularly for servers. It was developed in 1998 by a consortium of industry leaders, such as Intel. At that time, they were dealing with a world of 8-bit microcontrollers acting as service processors. These interfaces were portals to their out-of-band management infrastructure, but everything was proprietary; different vendors were using their own homegrown protocols and interfaces. And they realized, even then, that they needed a better way.
So, they created a specification called the Intelligent Platform Management Interface, or IPMI. And it’s done well in the industry. But we are now reaching a point where IPMI is stale. It can’t keep pace with changing infrastructure, particularly in hyperscale data centers. Look at the infrastructure advancements being driven by Open Compute, aka OCP, for example. Today, you have issues around scaling, and difficulty managing security threats – keep in mind the security threats are much more sophisticated today.
IPMI wasn’t really developed from the ground up with security in mind; it was kind of an afterthought. And there’s also a high barrier to entry. It’s not easy to pick up IPMI and run with it without a technical background.
AMAX: So this is where Redfish comes in.
Emerson Agustin: Yes! Redfish is really that next step in creating an open, easy-to-access hardware management interface for the industry. The industry leaders, including Emerson, realized that there were issues with the IPMI approach. IPMI was great, but we need something that matches more of the paradigms data center operators are using on the production side today, like web services and RESTful APIs.
We decided it was time to get together again and try to curate a new industry standard that creates a simple management interface, using a RESTful API. Under IPMI, the protocol, the way that you speak to the hardware, was intertwined with the way that the hardware behind that protocol was modeled, and that was a problem. It constrained you; every time you wanted to update your data model, or the way that you modeled devices in your hardware infrastructure, you had to make a change to the combined protocol and data model.
With Redfish, the two are separated so that you can define a standardized interface, and the data models are allowed to grow. IPMI wasn’t really architected well for extension, and the protocol became fragmented. With Redfish, you have extensions with guardrails. We applied the lessons learned from using IPMI over the last decade.
AMAX: The next question might have already been answered, but if Redfish is not backwards-compatible with IPMI, what will the architecture look like?
Emerson Agustin: First of all, Redfish is not backwards-compatible with IPMI from a protocol aspect. In terms of functionality, it’s in line with, and grows beyond, what IPMI can provide today, in terms of capabilities on the LAN interface. We have to be clear, because IPMI is used not just on the LAN or network-facing side, but it’s also used internally in the interface between the host processor and the BMC. So you’re going to have architectures, at least at first, which provide Redfish on the network-facing side of the out-of-band management interface. But they must also support, in tandem, IPMI and other protocols such as SNMP and CIM.
AMAX: Will Redfish offer the same level of management features that can make it comparable to current management systems?
Emerson Agustin: It will. Now again, there are things that Redfish doesn’t do today because it’s only on the LAN or network-facing side. But, my understanding is that they are being designed and planned for through the Distributed Management Task Force (DMTF), specifically the Scaled Platform Management Forum. So, Redfish should eventually support these other interfaces and functions. Beyond that, we, Emerson, would like to see Redfish extended to other data center infrastructure, not just servers.
AMAX: So what are potential roadblocks, and what are things that need to happen for Redfish to become widely adopted?
Emerson Agustin: One of the potential roadblocks is customers already have their toolsets that they’ve built up around IPMI, for example, or SNMP, or what have you. There has to be a clear migration path for them, otherwise it becomes too painful to adopt a new protocol or specification, and companies just won’t do it. They’re going to say, “I’ve invested all this time into my existing toolset. Why am I going to change now? Yeah, it’s a pain managing different sets of hardware, but if you guys can’t all agree on what this should do, I’m not gonna move.”
So, as industry leaders, we’ve got to make sure that we continue to do the good work that we’ve done until now. We released the 1.0 specification through the DMTF…now, we have to push mindshare. We have to educate the community at large, and tell them not just what Redfish is all about, but provide them with toolkits, examples, things that will get them onboard and moving in that direction. And then there will be a transitional period where we’re supporting multiple protocols on the network end, and Redfish. The adoption of Redfish will grow, and over time, the use of these older toolsets will ultimately fall off.
AMAX: Do you guys currently have any toolsets? So if software companies who are asking for these clear migration plans, do you guys have any kind of references yet?
Emerson Agustin: I believe this was part of the charter for the DMTF group to create frameworks and references. I know, prior to becoming an official effort under the DMTF banner, when it was just Emerson and its partners, we did create a schema model that allowed you to basically simulate what it would be like to talk to a Redfish-enabled server. So, you can expect to see more of this. Going beyond that, I expect coordinated efforts through DMTF, as well as individual efforts from the partners and Emerson to create toolkits and SDKs around Redfish.
AMAX: How does Emerson Network Power actually plan to put Redfish on the management system, on the power shelf itself?
Emerson Peter: Some people look at Emerson Network Power and think – “Why are they involved in this server management thing in the first place?” Emerson is one of the largest suppliers of embedded firmware for servers, for manageability with IPMI in the market. We supply roughly 50 to 60% of all shipping servers IPMI implementations, in whole or in part.
Emerson Agustin: When you think of Emerson Network Power, you don’t immediately say, “Hey, these guys do embedded firmware management for servers and other devices.” But it’s a big part of what we do. We’ve been involved in this community for over a decade. So, we are actively driving Redfish adoption in our solutions across infrastructure management firmware and software.
AMAX: Now assuming Redfish is the probable next step to IPMI, are all the BMC firmware providers gonna be on board with this? Considering there are quite a few different x86 servers on the market right now, are there plans for the other vendors to start using Redfish?
Emerson Agustin: This is a two-pronged question: there are server vendors and then there are different BMC firmware vendors, like Emerson. I’m not going to speak for the other BMC firmware vendors, but I can tell you that we’re obviously on board. In terms of server vendors, you can look at the DMTF charter to see who’s involved and you’re going to see a large representation of that community. So, given the amount of people that are actively participating in this effort, those who are monitoring it, or shown keen interest in it, I see broad adoption over time. The caveat is we’ve got to continue to do our job and push this forward. It’s a great idea, it’s got legs, and it needs to happen. We just need to follow through, and continue educating the public on what Redfish is and why it’s the right approach.
AMAX: How about the OCP community, are they on board with this management, since you basically have a power shelf that’s gonna be using Redfish, and it’s gonna have OCP compute nodes inside the rack, and things like that which need to be compatible. Where do the OCP ODMs stand in regards to the implementation of Redfish?
Emerson Agustin: I can’t speak to what the ODMs think about Redfish. But I can say, I think my last answer still applies. I can tell you that, as we’ve worked through this, we’ve stayed in touch with OCP as a community. Some of our members in the initial alliance, prior to the DMTF, and even in the DMTF governing committees now, have strong relationships with the OCP group. So they’ve been kept in the loop.
Emerson Peter: I know there are more and more workshops between OCP and the DMTF community. And when you look at the people that are very big into sponsoring the OCP: Hardware Management Group, and you look at the membership that’s driving Redfish, you’ll see a lot of overlap in the companies that are promoting both of those. So, is there going to be an adoption of Redfish by OCP? I don’t know. But there is a high likelihood that they’ll consider it. And does Redfish probably currently have frontrunner status? In my opinion, I think so.
AMAX: So then how does AMAX, as a rack-level solution provider of both traditional and OCP platforms, leverage this new technology for our customers that want to take advantage of white-box cost savings and flexibility, but still need the features of Tier-1 type data center?
Emerson Agustin: Well, we’re a little bit biased, but the first step would be to use white boxes that have service processors running our firmware. Next, Redfish should scale up the rack. So at the top level, you may have an aggregator or gateway that speaks Redfish; aggregating all the devices below it that may not. You can also specify a controller inside your power system or power shelf that adds zone management, for example, and speaks Redfish.
That would be one way to do it. But why stop there? In our opinion, everything inside the rack, and the data center, should be Redfish-enabled. Then, you can come over the top with toolsets like Trellis™ software that can provide that Tier-1 class management regardless of hardware.
AMAX: We actually have looked at the specifications and the schema involved, and it looks like RESTful API modeling is being adopted as today’s type of programming language.
Emerson Agustin: Exactly. We were talking earlier about the differences with, and some of the issues around, IPMI. We talked about barriers to entry, right? What you just said, is a direct answer to that. Because it’s a RESTful web service, it should be very familiar to people in the data center industry. It should be fairly easy to write Python scripts to talk to Redfish enabled devices. The point is always about making the customer’s life easier. And almost any pain point we address is going to be around simplifying the process for them, and making it easy. That’s what we’re trying to do with Redfish. We’re trying to make management easy and elegant, yet also powerful and effective.
AMAX: Well, besides the firmware update/BIOS update type of feature set, how about things that are actually used in IPMI? Like Serial over LAN, KVM over IP, things of that nature. Will they be supported as well?
Emerson Donnie: These are actually separate interfaces and distinct from IPMI over LAN. I don’t think Redfish is in a position yet to replace the Serial over LAN function. Another example is KVM over IP- it’s one of our specialties at Emerson Network Power, actually. But it’s not really something that’s in the same wheelhouse as IPMI- at least not directly. For example, you can use Redfish to configure the KVM session parameters, but you’re going to launch KVM through a separate interface like a dedicated HTML5 client UI. Now, other things done today using IPMI can be done by Redfish easily. The bread and butter of any management system, for example, is being able to turn something on and off.
AMAX: And sensor data, right?
Emerson Donnie: Yes, pulling in telemetry is part of it. So it’s really about aggregation, monitoring and control. And this goes back to your initial question around “what is server management;” that’s a big part of it. Pulling sensor data, getting nameplate info from the server, discovering it, powering it up and down and doing administrative tasks…the list goes on and on. These are management functions that should be easy to do, and are not necessarily easy with IPMI anymore, despite its best intentions.
This is what’s at the heart of Redfish, and this is where we believe the market is headed. We’re really excited about where we see things going, and we believe that the changes we’ve talked about in this conversation are going to begin happening in a rapid manner very shortly.
What are your thoughts on the potential of Redfish, and its impact on the future of the data center? For more on Redfish and other groundbreaking developments in the data center, cloud and HPC markets, tune in for future posts on AMAX’s blog!