For Entry IBM Shops, Power9 Is About Performance And Security
December 10, 2018 Timothy Prickett Morgan
Buying new systems costs money, often a lot of money relative to the size of the overall IT budget and the revenue and profit streams of the companies for which they work and, in essence, actually embody what that company really is. So in a sense, systems are always worth the money if they are actually letting people do their work properly.
That said, there is always an argument to be made for doing an upgrade – often actually a migration because the system itself cannot easily or economically be upgraded – and another set of arguments for waiting a little while longer because the current system is still running just fine, thank you very much. The argument is even strengthened, not weakened, if the current system has been in place for four or five years and has been acquired outright from the beginning or bought for pennies on the dollar after a three-year lease expires. If the machine is not costing any direct costs in terms of capital outlays, it sure does feel free, even if the maintenance for hardware and software is mounting up.
For customers using midrange and enterprise class Power Systems servers, it is not hard to come up with a strategy that takes advantage of the performance in each Power9 core relative to the performance of Power8, Power7+, or Power7 processors, and it is a snap to make the case for even older systems based on Power6+, Power6, Power5+, and Power5 machines that are even older. The Power9 core has roughly 50 percent more oomph than a Power8 core, 2X that of a Power7+ core, and 2.5X that of the Power7 core. So companies with lots of cores in machines in the P30, P40, and P50 IBM i software tiers can save a bundle on per-core software charges as well as lower maintenance fees for hardware and software by downshifting to systems with fewer cores. We talked all about that in last week’s issue of The Four Hundred for enterprise-class iron, and separately also had a chat about the nature of the current installed base and the issues of migration and upgrades with Steve Sibley, vice president and offering manager of Cognitive Systems at IBM.
We did a follow-up interview with Sibley and Simon Porstendorfer, who is in charge of the scale-out Power Systems, to drill down into the entry market. The machines we call entry systems when we talk about IBM i can be used as standalone boxes or they can be elements in a compute cluster that runs Linux or AIX workloads.
IBM did a survey in May of this year, three months after the scale-out Power Systems were launched, to try to figure out what customer concerns were and the reasons why they either had moved to Power9 iron or were planning to. The results are very interesting, as you can see below:
“Before the survey, if you would have asked me, I would have guessed that lower cost would have been a big factor,” says Porstendorfer. “But when I looked at the survey results, the migrations to Power9 are really about performance and security. Spectre/Meltdown has actually helped us, and we have been very open about this and the performance impact with security patches turned on. Not all of our competitors have done this, or have told customers what the impact of the Spectre/Meltdown patches is.”
For customers who have machines with one or two Power cores activated in their boxes, there is really not much of a core consolidation story, as is the case and as Sibley talked to us about last week. If you have two Power7+ cores, you can consolidate down to two Power9 cores and have about the same performance as gauged by the Commercial Performance Workload (CPW) benchmark test that Big Blue uses for IBM i database applications. Yes, you could downshift from two cores to one and have about the same performance, but what does that really get you? If you have a Power7 machine with two cores, you would end up with maybe 25 percent more performance and you might save some dough on having to license a second core. But those savings are not much – call it a few thousand dollars – out of the cost of a typical entry IBM i server, which is somewhere around $25,000 for the base system including operating system transfers from old to new machines.
You can also make a consolidation play at the level of whole machines, and the numbers sort of make sense if you use IBM’s sizing and recommendation system, which pumped out this chart showing the total costs of operating a pair of old Power6-based Power 520s for five years compared to a single Power S914 machine:
It costs close to $160,000 to operate that pair of Power 520 machines, if you keep them on hardware and software maintenance (which many real customers don’t do), compared to just south of $120,000. Call it somewhere around $40,000 in “savings.” But realize, too, that many customers want two machines for high availability clusters, so this consolidation argument kinda shoots that in the foot.
Hence the focus on getting modern and getting a lot more performance for probably about the same money that customers spent on machines four, five, or six years ago – and in some cases, even longer than that. This is just the reality of the situation.
IBM is seeing good activity, particularly with the Power S914, almost all of which are sold running IBM i. The Power S924 machine is more popular among AIX customers and is used as a standalone DB2 or Oracle database server running ERP applications. AIX shops tend to have bigger workloads, we presume.
“I think we are seeing some response, and I don’t know if it was the noise around the Spectre/Meltdown security vulnerabilities or the fact that we have been more verbal about our end of marketing and end of services plans, but we are seeing more people thinking about getting current,” says Sibley. “As you have pointed out before, it is actually economically inefficient to stay on old hardware and, more than that, you are putting yourself at risk. The older the stuff is, the more likely it is to break and the harder it is to find parts to service it. If you run a system long enough, even as good as we think we build it, it will fail at some point in the future. And there will be security vulnerabilities, and we are not going back to, for instance, Power5 or Power6 machines to fix the vulnerabilities in systems where we don’t update the firmware anymore. It pays for customers to stay current. They can stay longer – five years or six years if they want – but staying fresh on a technology basis is in their best interests, and hopefully they are realizing that at the same time they see new technologies and the path to advancement that is just as important as risk mitigation.”