“The scuttlebutt is that IBM seemed perfectly content to wait until May to launch the Power7-based Power Systems servers, but something changed and compelled the company to move up the announcement of its first machines using the eight-core processor to today. Big Blue is not in a habit of explaining its motives or its timing for product launches, but it seems clear that IBM wanted to get out in front of a whole lot of processor and systems launches that are expected between now and the summer.”
Looking over the specs of the machines released today, all I can say is wow!
I’m not surprised IBM has opted to release the mid-range and enterprise classes this quickly, their timing is impeccable. With the performance differences between POWER, SPARC and Itanium growing ever more steadily and still no clear idea when SPARC/Solaris will be back to “Business as usual”, now is indeed the time.
I finished negotiating a new job contract today where I was told by a traditionally Solaris house that my most interesting skills where X86/Solaris and AIX/Linux on POWER. When I asked them why they felt that way, they said that not only where Intel based Solaris machines faster than SPARC, but that POWER was by far the most interesting platform for high-end UNIX systems currently on the market.
Never in my life has being a systems and OS polyglot paid off so well 😀
I’ve always got the feeling that the support of x86 is going to be taken to a whole new level as Oracle realises that the days of SPARC are pretty much numbered outside of a few niche areas. As processors become more complex so will the costs of R&D will increase so unless you have the economies of scale to spread those additional costs over more units it’ll be a costly venture of either the CPU being behind the leader or so costly that the parent company never recoups the costs over the lifecycle of the processor (made worse now with the quick turn around with each design due to Intels tick-tock development model).
I do hope, however, that we’ll see some freaky chipset designs that really push the envelop when it comes to scalability, reliability and speed which is where Sun/Oracle can carve out a niche. Take a bog standard Xeon processor, attach it to a custom Oracle/Sun chipset and push Solaris scalability to the max.
As for OpenSolaris itself; they need more programmers – so many things are missing not only high end but also for the development desktop; it is time for Oracle to bite the bullet and realise that they’ll need to invest 4 years worth of profits into OpenSolaris to catch up to Windows and Mac OS X in the area of usability. Does it matter? if your desire is for a top notch development environment, a top notch Sun Ray/Thin Client experience – yes, the user experience is important! as for ROI as some will claim – the difference is either staying afloat or sinking – dollars don’t enter the equation, only survival.
Edited 2010-02-08 23:58 UTC
What’s your opinion of linux on power in practice? The idea of giving IBM vendor lock-in on linux has always seemed weird to me, and though I understand that there could be some benefits, I wonder how it beats, say, ESX server + linux on Dell or IBM.
To be honest, I think it’s to be avoided even if you take the vendor lock-in out of the equation. You are much better off using AIX in my experience as it’s better integration with the platform becomes apparent very quickly.
For example software RAID mirroring can be a PITA as the boot partition for Linux on POWER has to be a PPC PReP partition with a Fat16 file system that cannot be added to a Linux software mirror. This basically means that every time you recompile or add a module to the kernel, you have to dd from your primary disk boot partition to the secondary, leaving the way wide open for human error (I’ve seen a guy dd to the wrong partition and then need to reconstruct the mirror from scratch).
I’ve heard people try to argue about the open source software stack being better with Linux and although that may have been the case in the past, it certainly isn’t any more, at least when talking about server applications. Pretty much every FLOSS server project of interest has been ported since V5L. AIX provides a similar enough environment to Linux for porting to be trivial (the L in the version number stands for Linux).
Although this next analogy isn’t perfect, it’ll do for the moment. Running Linux as the sole OS on a POWER system makes about as much sense as buying an Intel Mac and replacing MacOSX with Windows7. You can get a system for half the price so why bother? On the other hand, partitioning the POWER system into two or more logical partitions using hardware virtualization and running Linux on one LPAR is a bit more like installing bootcamp, Parallels or VMWare Fusion and running Windows7 from there. If you absolutely want an LoP system running, that for me would be the better solution.
In the end, POWER systems are high I/O systems that are a tad overkill for most of the things you would use Linux for. All major DBs have been ported so the only time I can see this being a good idea is if you needed the high I/O and wanted to standardise your UNIX environments on Linux.
Very interesting, thanks. I guess the whole thing seems a little incongruous considering that commodity hardware had a big hand in making linux popular in the first place. But from what you say, it sounds like if you are already running a lot of AIX and happen to need a linux VM or two, it would make sense.
One thing I am curious to see is how Oracle and IBM will price databases on POWER7 cores. Per core licensing is about the only thing that makes me root for Microsoft (well, actually, I like SQL server a lot and wish they would port it to a better OS).
Absolutely. If you’re already running a large AIX setup, you probably already use POWER’s inherent HW virtualization so it would be trivial to set up an LoP VM or two.
You’ve got me stumped there. I’ve not had to deal with Licensing for a long time as I’ve general had a sales team to deal with all that. I do know that IBM tries very hard to undercut Oracle with DB2 on their own platforms, including aggressive per core licensing. So much so that last I heard, you couldn’t get a version of Oracle DB for zOS. Not sure if that’s still true though.
As for MSSQL, I’m not a DBA so I couldn’t say one way or the other how I’d feel about porting it to other platforms. I have heard some positive feedback from DBAs regarding MSSQL but I doubt very much it’s positioned in the right market segment for POWER. Possibly Linux on X86-64 but it’ll be a cold day in hell before we see that happening.
The machines reportedly run AIX or Linux.
http://www.pcworld.com/article/188790/ibm_launches_eightcore_power7…
High-end stuff. If you were to cobble together a small cluster of such machines, I would imagine you would fairly quickly arrive in supercomputer territory without necessarily attracting a massive power bill.
Edited 2010-02-09 05:48 UTC
That is one of the main reasons why POWER7 has been released such a short time after POWER6+. From the wikipedia article:
IBM basically won $244 million from DARPA to help develop the architecture.
… and unless they win another big DARPA or DoE contract, chances are that POWER7 is the end of the line for IBM’s PPC, at least as far as the high performance non-embedded space is concerned.
The cost of developing the architecture and keep up with their fabbing is too much for the relative small market for such systems, without having a larger market to subsidize the development like intel has, for example.
Last I heard, the only reason why IBM is still in the HW business is due to the services revenue it generates, once the bean counters in IBM decide the investment is not worth the return they’ll ax a lot of the cool tech that was applied from the PPC into the Z,P,I-series…and the associated ZOS, AIX and OS400 software stacks.
Although that sounds reasonable, it’s not at all what I’ve heard. There was an IBM symposium in 2009 in which they clearly showed their commitment to the POWER line with a progression from the then new POWER6 to the soon to come out POWER7 and onwards to a POWER8 architecture that is already in development. The following link has a video in which the whole thing is discussed. Unfortunately, the whole thing is not only in German, but in Swiss German to boot!
http://www-05.ibm.com/ch/events/symposium/
Still, their is a link to a PDF that is mostly in English with the clear progression shown on page 16.
http://www-05.ibm.com/ch/events/symposium/pdf/42_J_Kunz_IBM_Symposi…
I’m pretty sure that’s why IBM is trying to corner the high-end UNIX market, like what they’ve done with the mainframe. They’ve already off-loaded vast amounts of the development costs by using standardised interfaces like PCI-X but have decided to keep developing the processors in-house due to the crap that Motorola came out with the last time IBM contracted in a partner for POWER.
Thing is, IBM makes more money out of POWER than most people realize. When I was working in Luxembourg, banks where defecting to the POWER platform en-masse from Sun and HP simply due to the architecture being the only Big-Iron UNIX offering with that kind of performance, with the energy efficiency that is all the rage at the moment and a clear and precise upgrade path. So much of IBM’s storage division relies on POWER that to axe it, the effects of the dominoes that would fall down would be catastrophic.
For example, let’s take arguably one of their most profitable offerings, the Mainframe division. Not only are the zSeries processors based on POWER developments, the external SAN disk bays used by Mainframes, notable the DS6000 and DS8000 series, have one and two POWER systems integrated, running AIX for all processing tasks, respectively.
Speaking of storage, this is again one of IBM’s big sellers. One of the biggest selling points for the POWER architecture is it’s exceptional integration with storage networks. At the moment, IBM has only one serious contender for storage and that’s EMC. If IBM where to stop developing POWER, they could in theory contract out their high-end disk systems to EMC, like they’ve done with their mid-range DS3000 to DS5000 systems that are re-branded LSI disk systems. But IBM would lose out on licensing for such things as AIX integration and FICON, areas where EMC are currently playing catch up. They would also lose control of their much bragged about tight integration between those systems. Marketing would no doubt throw a fit over that one as it’s a huge stream of revenue.
Lastly, DARPA are highly likely to finance future development costs of POWER as IBM are not only willing to develop the architecture with DARPA requests in mind, they are willing to push those ideas to they’re logical maximum conclusion, something we see with IBM’s willingness to reduce power consumption by a factor of 4 and ratchet up performance by a factor of 2 when the request was for a factor of 2 and 1 respectively.
So for me, the whole debate about POWER being iced makes no sense. It’s the same as the whole “AIX will be axed in favour of Linux” story we where hearing about a few years back. The bean counters are seeing far to much knock-on revenue to axe either project. With POWER’s current market dominance, it’s doubtful we’ll see a change in their attitudes in the near to medium future.
The development of a high end processor and its associated fab technology will soon reach the billion dollar mark.
The issue is that the development costs are growing at a much faster rate that the high end market does. So eventually it is not an economically feasible proposition.
Now, IBM are smart to not shit where they drink from and won’t announce publicly anything but their utmost support for POWER. But a company would never lie, right…
Edited 2010-02-10 07:38 UTC
Lol! How true!
If it’s true that it’s approaching a billion dollars, I wonder what IBM are going to do? Moving to another platform just isn’t feasible at the mo and won’t be for a long time to come.
Maybe spin off the fab?
Yeah, it is one of those things that it will not done right away. But the writing is on the wall…
I assume IBM will try to spin off their microelectronics branch, probably sell it to Global Foundries (AMD and IBM had a pretty close fab partnership up to recently, so it makes all sorts of sense).
This is nothing new, it really has happened all through the past few decades. Companies make huge moves to new platforms once their development costs lead them to some dead end. And the lesson seems to be that proprietary (although everything really is proprietary when it comes to corporate products) approaches end up in dead ends. While solutions which can rely on production volume can subsidize the development costs and avoid said dead ends.
IBM actually did it by moving their entire mainframe and OS400 lines from their old 3×0 ISAs into the PPC ISA. I am sure they could do it once there is enough virtualization support, coupled with the required error correction/high reliability design elements in the commodity silicon. And it makes sense, since at the end of the day… the processor is not the main value proposition for their systems. The actual system, and its integration is.
So IBM will continue selling highly reliable, highly scalable turn key systems. The actual native ISA of which at the end of the day will be of little importance to the customer as long as their intended applications work without any headache and with a clear cost/performance proposition.
Amen to that, brother!