The Possibilities With IBM i Entry Systems Sporting Power8
September 23, 2013 Timothy Prickett Morgan
At the end of August, IBM‘s top techies from the Austin Power processor development labs showed off the features and functions of the forthcoming 12-core Power8 chip at the Hot Chips conference at Stanford University. As The Four Hundred pointed out in its coverage of the divulged specifications at the time, this processor will pack a serious punch. Perhaps way too much for most entry IBM i shops, in fact.
Or, perhaps not. Depending on how you want to make use of the substantial performance that Big Blue intends to cram into the Power8 chip, which is up and running in test systems in the lab and which is expected to be delivered sometime around the middle of 2014. In a presentation, an IBM engineer said that the Power8 core would have about 1.6 times the single-thread performance of the Power7 processor (presumably with the Power8 running at its target 4 GHz clock speed) and would have twice as many threads per core. (That’s eight for Power8 and four for the Power7 and Power7+ chips.) Those extra threads can help squeeze out extra work for software that likes threads, such as Java application servers and databases. Here is how IBM says the Power8 chip will stack up against the Power7+ chip, socket for socket, on various workloads:
Depending on the machine that the Power7+ chip is dropped into, the eight-core version running at 4 GHz would deliver about 61,500 units of performance on the Commercial Performance Workload (CPW) benchmark test that IBM uses to measure the relative speed of different configurations of its Power Systems iron running the IBM i operating system. Now, as you can see on the chart above, the 12-core Power8 chip will have somewhere right around 2.5 times the performance of the eight-core Power7+ chip. The increase in core count, the increased cache memory size and extra layers of cache memory, the doubled simultaneous multithreading, and the changes to the core all significantly boost the performance. (I looked at the chart above three times because I was not sure I believed it.) What this means, to bring it on home in practical terms, is that a Power 720 or Power 740 entry machine will have something on the order of 154,000 CPWs of raw processing capacity.
Let that sink in for a moment. This is a huge amount of capacity. Even if IBM gears it down to four or six cores, and clocks the chip down, you are still talking about a lot of oomph in a single socket. Which is all that a Power 720 machine has. A Power8-based Power 740 machine would have slightly more than twice as much aggregate performance, assuming the processors ran at a slightly higher clock speed. (Maybe 4.5 GHz or 4.75 GHz compared to 4 GHz in this theoretical future I am thinking about. IBM has not confirmed any clock speeds, except to show some feeds and speeds of the processor’s features at a target 4 GHz clock speed.)
And I have this totally crazy idea. I want IBM to make all of this capacity really inexpensive, and also make Power 720 and Power 740 machines with lots of memory and I/O capacity to try to encourage midrange customers to put more and more workloads on top of IBM i and its DB2 for i database.
This is not, obviously, how IBM creates the Power 720 that we know and love today, of course. The typical IBM i customer has from two to four cores running that operating system, and maybe you see some customers with six or eight cores. There are some big midrange and high-end shops running IBM i, of course, but most customers have a Power 720-class box with a single processor socket.
What IBM has done brilliantly with the mainframe is to have extra engines latent in the box and offer customers a way to use them to accelerate Java, XML, and database workloads or to run Linux on the relative cheap. I say relative because at something close to $100,000 per engine for both the motor and the Linux license, a mainframe core running Linux is anything but inexpensive. But it is most definitely a lot less expensive than paying for a mainframe engine running IBM’s z/OS operating system.
What IBM needs to do, however, to get customers to consume all of this capacity, with IBM i or Linux, is to shift from its traditional perpetual licensing to monthly subscription pricing. Just make it as painless as possible. Take the cost of the IBM i license, add on Software Maintenance for three years, and divide by 36. Now, for existing customers who agree to add workloads to their machines, cut that price significantly (maybe 50 percent) to encourage them to add capacity for new workloads.
The reason why IBM i shops have so much X86 iron running Windows and a relatively small amount of capacity running on IBM i on a pretty small machine is that for the past 15 years, IBM has charged too much for IBM i capacity relative to Windows platforms. I know all of the great things about IBM i and its integrated database and sophisticated workload management, but if customers decide to put new apps on Windows Server 2012 and SQL Server 2012 instead, then none of this matters. It is all about results.
I would also like to see some denser and interesting form factors. I want to see a single-wide Flex System chassis, as I have said that entry IBM i shops need, and while I am at it, I want to see single-socket and dual-socket server Power8 nodes running IBM i inside the NextScale chassis that IBM announced two weeks ago and will start shipping in October using Intel’s latest Xeon E5-2600 v2 processors. In fact, service providers building IBM i clouds should already be able to such Power-based nodes using the existing Power7+ processors. If IBM wants to sell entry Power Systems, it needs to have all of the form factors–rack servers, tower servers, Flex Systems in single-wide and double-wide enclosures, and NextScale vanity-free systems–supporting Power chips. IBM has to put its own chips out there right alongside of Intel’s chips as part of every announcement.
If IBM wants everyone to believe that it believes in Power, it has to start treating Power like a peer to Xeon. Every time. It really is that simple.
As for the Power 720-class machines in the Power8 generation, I hope that IBM boosts the memory and storage capacity of the machine. There will be a whole new I/O architecture with the Power8 chips, and IBM should realize that it is unreasonable to artificially crimp I/O capacity on the Power 720 just to try to peddle the Power 740. Some customers don’t need that much compute when they do need I/O and storage, and a smart system designer creates a machine that lets the customer decide which way to build it without sacrificing anything. Let processing, memory, and I/O scale up and only build one machine. Simplify the product line, simplify the sales cycle, and sell more boxes.
Next week, I will ponder what IBM can and should do with midrange and high-end Power8 systems with respect to IBM i shops.
If you have any ideas, please do share. Now is the time to tell IBM what we want.