Observations, Priorities, And Desires For IBM i In 2020
January 6, 2020 Timothy Prickett Morgan
With age and wisdom comes a certain kind of calm expectation that does not necessarily put one in the frame of mind to make prognostications as much as making some observations, laying out priorities, and expressing desires. Making predictions is really about placing qualified and intelligent bets, more like hunting than the slower, measured pace of agriculture, which is about planning and fostering growth.
As I approach the new decade, I am less in the mood to don my wizard hat with its moons and stars on it and to rub the dust off the crystal ball on my desk – it’s really more of a flat piece of quartz, and frankly the big point of smoky quartz I dug out of the ground in the red dirt of the Piedmont of western North Carolina works best – and more in the mood to just make some observations and perhaps express some heartfelt words of encouragement to all of us in the IBM i market.
The first thing I wanted to bring up is that this is the start of Volume 30 of The Four Hundred, and it is now the 32nd year that I have been working on an IBM midrange publication. That is a long time for any group of people to be engaged in a professional relationship – more of an extended conversation – and I am grateful, as all of us who work on The Four Hundred and have over its many incarnations, to the readers and vendors who support what we do. I am simultaneously amazed that we are all still doing this, and at the same time not surprised at all given the benefits of the platform. A good architecture is hard to beat and longevity is its reward. And we think that the platform has many good years ahead – and we are not foolish enough given the nature of the market and business computing to try to predict the day the world will end. The System/3 and its follow-ons down to the IBM i of this day have been around literally since 10 days after men first walked on the Moon with the Apollo 11 mission.
“That’s one small step for a man, and one giant leap for Mankind.” – What Neil Armstrong meant to say. . . .
Anyway, it’s a fool’s game to predict the demise of the IBM i platform. It is still a vibrant, if somewhat smaller, customer base and the shrink in the revenue stream is mostly due to Moore’s Law advancements in computing and storage than anything else. Seriously. The base has contracted by 2X in three decades after transforming from System/3X to AS/400 gear in the 1990s, but the amount of compute crammed into a system and the work it can do for a given dollar has gone down exponentially and represents most of the decline in revenue. Big companies needed a fleet of single-processor AS/400 B60s and B70s and prayed for D80s (two processors) and E95s (four processors) and eventually got Power 795s with 32 processors and we can now cram 12 cores in a socket instead of instead one and will probably be able to do twice that with the Power10 chip in 2021. (Those are many-threaded fat cores used with IBM i and AIX , not the skinny cores with half as many threads used on Linux machines based on Power CPUs.)
That brings up the next point. We don’t know exactly when IBM will launch the Power10 chip, but we know it is not going to be this year. The Power9′ chip, aimed at Linux shops and testing out a new memory subsystem that will be deployed with the Power10, is coming out this year, we think before the International Supercomputing conference in Germany in June, and we don’t expect to see the Power10 until maybe early 2021. At best. It could be the early summer. And that is a long time between Power processor generations, with the initial Power9 chips shipping in late 2017 for the Power AC922 nodes used in the “Summit” and “Sierra” supercomputers built by IBM for the U.S. Department of Energy. IBM i didn’t get Power9 chips until the spring of 2018, so the gap between IBM i processor generations will probably be around three years, which has been the average and which is still more or less on the Moore’s Law curve even if there are half as many steps between processors. There are not really Power “plus” variants any more. IBM does a process shrink and an architectural change at the same time with no interim stage.
This cadence meets the buying patterns of IT departments at midrange and large enterprises as much as it does the buying patterns of big supercomputing centers, so it is no accident. Hyperscalers and cloud builders, who make money from systems as opposed to using systems to count money they make doing other things like making or distributing products, tend to replace a third of their enormous fleets of millions of machines every year, and so they drive an annual cadence. These are radically different customers with very different needs, and the fact that there is not an annual cadence for Power chips aimed at hyperscalers and cloud builders shows just what IBM is really expecting to happen here. AMD and Intel and the Arm collective – at this point it comes down to Ampere and Marvell – all have annual cadences of CPU rollouts. If IBM wants to play here, it needs to do the same. With the memory bandwidth improvements coming with the Power9′ and Power10 processors, I think Big Blue has a real chance to break into the hyperscale and cloud markets because most workloads are not compute bound, but memory bandwidth bound and with all of the I/O that allows so many things to be hung off a Power chip, there is no better processor for streaming huge amounts of data from media to CPU and/or accelerators to be chewed on and dumped back into memory and then media. IBM will be widening this gap in 2020 and beyond, and there is no read why it can’t shift to an annual cadence to keep the innovations rolling.
The IBM i base could leverage all of this technology, which we described here, and in fact, coupled with the native NVM-Express flash support that came out with IBM i 7.4 Technology Refresh 1, which we talked about here, one could make an extremely fast Db2 for i database machine that would scream – just completely scream – on either batch or interactive work. What would you pay for a machine that didn’t have a batch window because it was just that fast? Would you upgrade now?
It may be possible to make a machine so fast that you don’t need to do GPU acceleration, but all the same, we still want IBM i to run on a GPU-enhanced machine like the Power AC922 because that architecture is used with traditional HPC and new machine learning workloads. There is no reason why IBM i, the integrated platform, should not be running the simulation and modeling programs for making products or running risk analysis for financial transactions or routing products in distribution networks as well as creating the recommendation engines using machine learning algorithms. It should all be on one box, and I think it should be a variant of the Power AC922 equipped with the Power9′ processor that has a couple of Nvidia Tesla V100 GPU accelerators in it.
Let’s bring all of these technologies together and create a new IBM i platform for the new decade. We don’t have to wait for Power10. We can get started with Power9′ now, and be ready to fully exploit Power10 when it gets here somewhere between a year and two from now.