IBM has been making a lot of noise in the datacenter recently. In December of 2017, the company’s POWER9 platforms hit the market with a focus on Artificial Intelligence (AI). In March of this year, Rackspace and Google gave the first public peak at the Zaius and Barreleye 2 Open Compute platform at the Open Compute Summit. At last week’s Red Hat summit, IBM and Red Hat announced a partnership to extend IBM Cloud to Red Hat's OpenShift Container Platform and Certified Containers.
These are all significant, but perhaps the biggest bang of all is the delivery of IBM and Nutanix hyperconverged platforms based on IBM’s POWER platform. So what exactly is this, and what does it mean?
First, a POWER9 refresher
My firm’s Principal Analyst, Patrick Moorhead, wrote about the POWER9 CPU
at product launch. One of the product’s key differentiators is its richness of I/O. PCIe G4, NVLink
, and OpenCAPI
are all enabled in the POWER9 architecture, which enables Graphics Processing Units (GPUs), field programmable gate arrays (FPGAs), networking and memory class storage. All of this adds up to a processor designed for AI at its core. In fact, IBM contends POWER9-based servers can increase Machine Learning speeds by up to 4x.
That being said, AI is not the only workload that can benefit from the POWER9 architecture. IBM has taken the POWER9 architecture and developed systems tuned for four different usage models:
What exactly did IBM and Nutanix announce?
- Accelerated (AI/ML, Analytics, HPC)
- Enterprise (Cloud, virtualized infrastructure, ERP)
- Scale Out (cloud, mission-critical apps and database environments)
IBM and Nutanix have partnered to deliver the industry’s first hyperconverged platform on the POWER architecture. Using IBM’s scale out servers as the underlying platform, IBM is shipping two hyperconverged platforms:
Why this matters for enterprise customers
- CS821 is a single node system based on the IBM S821LC platform. This 1 rack unit high server is designed for high-density virtualization and middleware. It’s also a solid building block for private cloud.
- CS822 is also a single node system based on the IBM S822LC platform. This 2 rack unit high server is aimed at higher performance workloads such as enterprise databases and tier one line of business (LoB) applications. While the CS822 has slightly more compute power, memory and storage are what makes this hyperconverged solution so powerful.
How does all of this work? AIX
is a UNIX operating system developed and sold by IBM. Many enterprise IT organizations have designed, built, and maintained tier one LoB applications in IBM’s Db2
database and WebSphere
application server. Because of the mission-critical nature of these applications, IT organizations cannot afford to take them offline for any amount of time. As a result, these applications tend to receive incremental updating and maintenance. While interfaces may change and evolve, that three-tiered application built on a Db2 database and WebSphere application server still runs on a POWER-based server running AIX.
In today’s IT organizations, these applications reside on separate hardware that must be managed and maintained separately from x86-based Linux workloads and applications. By deploying IBM’s hyperconverged servers, an IT organization can now run that AIX-based LoB application on the same hardware as newer workloads running in Linux environments. This can lead to real TCO savings through server consolidation and “single pane of glass” management.
The real value of IBM HyperConverged Systems
“Where simplicity meets performance” is a tagline that IBM seems to be using across a lot of its marketing material. I believe this sums up the IBM HCI story perfectly. By partnering with Nutanix, IBM is greatly simplifying the deployment, management, and maintenance of its POWER9 platforms and the workloads that run on them.
I will be curious to see how this partnership plays out over time. Is an Artificial Intelligence/Machine Learning HCI offering somewhere off on the horizon?
Will this spark a POWER revolution in the datacenter?
IBM’s initial approach to the market appears to be with its existing customer base—organizations that require dedicated resources to manage and maintain two separate infrastructures, operating systems, and applications. This is a smart strategy since this audience deals with this infrastructure management headache on a daily basis.
As the POWER9 architecture has an affinity for AI, analytics, and HPC, I would expect IBM to attack these markets next. These workloads tend to be deployed and supported by IT organizations with deeper roots; the performance capabilities of an HCI solution based on POWER9 could hold strong appeal for this audience. I don’t believe IBM’s HyperConverged offering is going to supplant x86-based HCI platforms, but it should certainly extend IBM’s reach.
Signs of a bigger, quieter revolution
While I don’t believe Intel
’s presence in the datacenter is threatened to any real extent, there is a growing POWER presence in the server market. Rackspace and Google’s collaborative Open Compute/OpenPOWER Zaius/Barrelye2 project that I mentioned earlier is an effort to bring an open-sourced POWER server design to the hyperscale community. This is the first non-IBM effort I’ve seen where OpenCAPI is fully built into the motherboard design, along with NVLink. Google has embraced and deployed this platform because of its core and thread density (Patrick Moorhead has covered this deployment in depth here
It will be interesting to watch the adoption curve of POWER in hyperscalers and enterprise customers alike. I believe IBM’s HyperConverged play is a smart attempt to gain greater traction with existing enterprise customers (and perhaps those IT shops that have been considering a POWER-based analytics solution, but lacked the resources to deploy and manage). More interesting to see will be IBM’s transition to the broader enterprise IT market—I will wait to offer opinions on that effort at a later date.