The Platform Matters More Than Ever, The Operating System Less So
April 16, 2018 Timothy Prickett Morgan
Those of us who have been in the IBM midrange for three or four decades are so used to thinking about systems and application platforms for so long it is hard to remember that this sort of thinking is fairly new to the rest of the IT community that did not grow up on proprietary mainframes or midrange gear. Platforms matter more than ever and continue to evolve, but that core operating system – the basic kernel that runs on a processor and has some management features on it – is changing very slowly.
This is not a bad thing. Or a good thing. It is just a thing.
I got to thinking about this recently while analyzing Microsoft’s recent tech preview for its Windows Server 2019 operating system, which will start shipping sometime in the second half of this year. That tech preview followed in the wake of the mostly regular Technology Refreshes for IBM i 7.2 and 7.3 announced in February and shipping in mid-March when Microsoft started talking about the forthcoming Windows Server. I think a lot about platforms in our day-to-day working life, because that is the interesting bit of information technology, taking a holistic view of everything from the elements of a core in a processor, out through the cache and memory hierarchy, to storage and I/O subsystems, up through the kernel and systems software, to the compilers and the applications they form to fit to the iron, and finally out to the screen where end users reach in and make these bits and bytes wiggle and jump.
Honestly, it is only through incremental improvement that any of this works at all. Which is why we see so much incremental improvements these days and not monumental change. I sometimes miss the days of monumental change, in that Cambrian explosion period a few decades ago when there was maybe two dozen different processor architectures and maybe two or three times as many operating systems – and we mean relatively lean operating systems here – running on top of them. The fast pace of change for hardware and software was exhilarating, and there was a growing pile of money every year that system makers were fighting over. But over the decades, the best minds in the industry have advanced the state of the art and have done the zillions of things from the core of the kernel on out to the application that, for the most part, gives each and every company that wants it a stable platform on which to build applications. In a sense, all operating systems are mature, so they are not changing very fast anymore. Things change very slowly, and that means things are stable – at least at the level of the operating system, the relational database, and the middleware that runs the applications.
Windows Server 2019 is a case in point, and going through the highlights shows it. Back in the day, when a new Windows Server release came out, everyone was obsessed about its scalability and reliability and how it compared to alternatives such IBM i, a slew of Unix variants (including IBM’s own AIX), and the IBM mainframe platforms: VSE, OS/390, and VM. We all dug through the manuals to see how many processors or cores or threads it could span, how much memory it could address, what the impact of SMP or NUMA clustering was on performance, how the I/O was architected to match whatever new gizmos were on the PCI-Express bus. No one really worries about these things. It is a given that any operating system will exploit hardware, and that most hardware is more than enough for most customers. This is not just an IBM i thing. Certain customers, to be sure, can make use of as many cores Intel can cram into a two-socket server, but for most companies, they are nowhere near the top bin parts and they have much less capable processors running at a lot lower cost and with plenty of excess capacity. It doesn’t matter if it is Windows Server or Linux. The basic workhorse server does not look that different from a Power8 or Power9 machine, and in many IBM i shops, there is far less compute dedicated to IBM i on a single instance than on a Windows Server or Linux machine. The database jobs that most IBM i shops have are fairly modest.
One of the big changes with Windows Server 2019 that the operating system has a new management console, called Project Honolulu, that takes the metaphorical interface of the Azure public cloud and its virtualized and containerized instances and the services that work for it and brings it into the on-premises Windows environment. The idea is to integrate Azure backup, disaster recovery, database, and compute servers with the on-premises applications, and this has to be done at the management level and can also be done higher up at the platform level using the Azure Stack software Microsoft has created to run private clouds. The regular Systems Center management console will also be updated and is not being deprecated. Yet.
On the security front, Windows Server 2019 will have an embedded version of Windows Defender Advanced Threat Protection, a server variant of the virus and malware killing software that comes with Windows 10 on desktops. The operating system will also see Linux virtual machines supported alongside Windows VMs in the Shielded VMs that were launched with Windows Server 2016. Windows Server 2019 will get an improved Windows Subsystem on Linux, which allows for Linux containers to run side-by-side with Windows containers, and Microsoft is, of course, working on a variant of Kubernetes container orchestration for this hybrid Windows-Linux environment.
Finally, Windows Server 2019 will cost the same as Windows Server 2016 and come in the same editions, but Microsoft warns that “it is highly likely” that it will raise the price on Client Access Licenses (CALs) for access to the server. (This is a different Client Access from the one on OS/400 and IBM i.)
See what I mean?
Above the operating system, no matter which one, there is a slew of stuff that is changing, some of it exceedingly fast. The whole container infrastructure stack – which gives these monolithic operating systems like Linux and Windows Server something akin to the subsystems that you are familiar with for decades now – is one. The cloud and virtualization stack, which was all the rage a decade ago, has itself matured. No one is counting VMs per hypervisor or virtual memory or virtual disk capacity. IBM got the KVM hypervisor and the OpenStack cloud controller to Power chips, but KVM is not able to run side-by-side with IBM i or AIX on machines, and OpenStack is not really IBM i friendly or pushed by Big Blue. IBM is not doing anything to integrate Kubernetes with IBM i. Microsoft is trying to keep Windows Server relevant to Linux because it has to keep the Azure cloud appropriate for Linux, and that whole stack of container infrastructure, clustered and hyperconverged storage, and data analytics and machine learning applications that the open source community is writing like crazy has to work on Azure more than it has to work on Windows. In a sense, Linux is like the PASE AIX runtime in IBM i. Microsoft has to put it in there to stay relevant, and in many ways it is helping it to not have to reinvent so many wheels.
It would be interesting to see IBM make use of all of these new technologies to create a variant of IBM i that has all of the new metaphors for compute, networking, and storage, that made it relevant to new classes of users. We desire this. Even still, we would love to see a whole new unified operating system from IBM that had runtimes that let it finally converge mainframe, Linux, Unix, Windows Server, and IBM i workloads on a single platform. But that is an academic pipe dream, and one without much economic justification. We just want IBM to create OS/500 or IBM i2 or whatever you want to call it, just for the fun of it and to show how to make a truly integrated modern system.