blockchain and the mainframe
Blockchain

LinuxONE Emperor II- the Linux Mainframe

The latest in the LinuxOne lineup is the LinuxONE Emperor II, introduced this past summer. It is built on the z14 and it runs a dedicated, optimized version of open Linux. It can do everything you can do with a native (x86) Linux system and more. You can run Docker containers to your heart’s content. Spin up dozens, hundreds, even thousands of Hadoop instances if you want massively scaled out Hadoop analysis. You can run it with Spark for cognitive analytics.

Unmatched scalability and processing efficiency

Maybe the primary use case for the machine is massive scalability. It can serve up to 30 billion web data requests a day. Furthermore, with its high consolidation ratios, the virtualization capabilities in a single Emperor II system can result in a less complex infrastructure with fewer components, less management, less space requirements, and lower software costs than x86 servers.

For example, it provides the capacity to do the work of thousands of x86 cores, cores for which you would otherwise have to pay a software license fee if you were deploying multiple x86 cores. Along the same lines, you can run up to two million Docker containers in a single system. Did we say processing efficiency; this is processing efficiency on steroids.

Multi-dimensional growth and scalability

The machine can be packed with up to 170 cores, up to 10 TB of memory, and up to 160 PCIe slots. And you can use this capacity in a variety of ways. You can add more system resources—more cores—to the service of an existing Linux instance or clone more Linux instances with a high degree of resource sharing. That’s scale-out capabilities in excess of anything you can achieve in the x86 world and capabilities you can execute with just a few keystrokes. For example, you could, as IBM puts it:

  • Dynamically add cores, memory, I/O adapters, devices, and network cards without disruption.
  • Grow horizontally by adding Linux instances or grow vertically by adding resources (memory, cores, slots) to existing Linux guests that may have initially been under-configured.
  • Provision for peak utilization and, after the peak subsides, return unused resources automatically to the resource pool to be reallocated for a different workload. And, of course, you can reallocate from the same resource pool should another peak arise.

LinuxONE vs. x86 TCO

There is no doubt that IBM is targeting the x86 (Intel) platform with its LinuxONE lineup and especially its newest machine, the Emperor II. For example, IBM reports it can scale a single MongoDB database to 17TB on the Emperor II while running it at scale with less than 1ms response time. That will save up to 37% compared to x86 on a 3-year total cost of ownership (TCO) analysis.

The TCO analysis gets even better when you are looking at priced-per-core data serving infrastructures. IBM reports it can consolidate thousands of x86 cores on a single LinuxONE server and reduce costs by up to 40%.

Software is the culprit

According to IBM’s competitive cost analysts, software costs typically make up the largest percentage of TCO. Are you surprised? Most managers of x86-based shops are confident that the low cost of x86 machine ensures a low TCO.  So confident, in fact, that they often don’t bother to add up the other costs in the full TCO analysis. Software—often licensed per core—is the largest single cost, but other costs include the people involved, network and storage costs, and facility costs, which can include the cost of providing electrical service to those cores and cooling to remove the heat those cores generate.

Software becomes the highest TCO cost because many software products are sold by the core.  Read the fine print on your software licensing agreement: Typically, the amount of charged cores is usually determined by the server’s maximum number of physical cores, whether or not they are activated or used by the software. And then some architectures require more cores per workload.

Finally, x86 to IBM Z core ratios differ per workload but x86 will invariably require more cores than enterprise servers. Remember, the LinuxONE is a Z System. For example, the same WebSphere workload on x86 that requires 10 – 12 cores may require only one Integrated Facility for Linux (IFL) or less on the Z. So the lesson here: whether you’re talking about system software or middleware, it will have an impact on TCO, which you need to consider.

Compelling Real-world Example

IBM reports that a North American health insurance company deployed LinuxONE over x86 for instant provisioning and metered usage. Here’s why:

The insurer needed fast and flexible provisioning for its database workloads. The company’s approach directed it to deploy more x86 servers to address growth. Unfortunately, the management of software for all those cores had become time consuming and costly. The company deployed 32 x86 servers with 768 cores running 384 competitor’s database licenses.

By turning to elastic pricing on the LinuxONE Emperor II it generated serious savings. The LinuxOne Emperor used 63 IFLs running 64 competitor’s database licenses.  It estimated savings of $15.6 million over 5 years just by eliminating charges for unused cores. (Full disclosure: these figures are provided by IBM; the author did not interview the insurer to validate this data. Also, please note, there are many variables at play here around workloads, usage patterns, labor costs, and more. As IBM warns: Your results may vary.)

The lesson here is simple. There is more to cost than the acquisition price of the server. That low cost x86 machine may cost you millions more than you realized just in the software cost for each core. At least do the full TCO analysis. You may be surprised to find that the LinuxONE turns out to be the less costly choice in the long term.

Alan Radding

Alan Radding

DancingDinosaur is Alan Radding, a veteran information technology analyst, writer, and ghostwriter. Please follow DancingDinosaur on Twitter, @mainframeblog. See more of his IT writing at technologywriter.com and here.
Alan Radding

Leave a Reply

Your email address will not be published. Required fields are marked *