IBM Introduces the Commodity Power Server
-
The S812LC is a one-socket, 2U system, equipped with up to 10 processor cores, 1TB of memory, 115GB/s memory bandwidth, and up to 14 disk drives. It starts at under US$7,000, though a fully loaded system will be considerably more.
2U 14 drives - 2.5" for sure
-
-
The real question... can the NTG Lab get one of these donated?
-
I wonder if you could install XenServer (or just use Xen) or KVM on one of those... Load it up with 1TB of ram... Yeah, I'd go for that, lol... As long as somebody else paid for it, lol.
-
How trivial is it to convert x86 Linux applications to the Power RISC architecture? Will it make sense for developers to devote resources to that process? Or are we going to see this used more for custom applications in the enterprise?
-
@dafyre said:
I wonder if you could install XenServer (or just use Xen) or KVM on one of those... Load it up with 1TB of ram... Yeah, I'd go for that, lol... As long as somebody else paid for it, lol.
Nope, neither exists for Power yet. Xen has been ported to ARM, so in theory probably trivial. Virtualization should be native to the hardware. Power has always had virtualization right on the hardware.
Don't forget that you get containers too.
-
@coliver said:
How trivial is it to convert x86 Linux applications to the Power RISC architecture? Will it make sense for developers to devote resources to that process? Or are we going to see this used more for custom applications in the enterprise?
Totally depends. If you were writing on Python, Ruby, PHP, Java, Scala, Clojure, etc. then there is nothing to port. Only when you write in C, C++ or similar is it an issue. In many cases, pretty trivial as it goes since Power is biendian.
-
Red Hat, Suse and Ubuntu have already ported their entire ecosystems. How many things are actually written just for AMD64 architecture anyway?
-
@scottalanmiller said:
@coliver said:
How trivial is it to convert x86 Linux applications to the Power RISC architecture? Will it make sense for developers to devote resources to that process? Or are we going to see this used more for custom applications in the enterprise?
Totally depends. If you were writing on Python, Ruby, PHP, Java, Scala, Clojure, etc. then there is nothing to port. Only when you write in C, C++ or similar is it an issue. In many cases, pretty trivial as it goes since Power is biendian.
That makes sense. They will probably have to be tweaked to take advantage of some of Power's components though? Or does the underlying language handle all of that?
-
@coliver said:
That makes sense. They will probably have to be tweaked to take advantage of some of Power's components though? Or does the underlying language handle all of that?
What do you mean?
-
@scottalanmiller said:
@coliver said:
That makes sense. They will probably have to be tweaked to take advantage of some of Power's components though? Or does the underlying language handle all of that?
What do you mean?
Language was the wrong choice of words, the article answered my question... probably should read before I comment :).
-
@scottalanmiller said:
Red Hat, Suse and Ubuntu have already ported their entire ecosystems. How many things are actually written just for AMD64 architecture anyway?
Knew about SuSE and kind of about Ubuntu, but didn't know RedHat had a PPC port. Though most who do use that specific kernel it would be using Yellow Dog most of the time. PPC was kind of the forgotten platform for Linux, right before SPARC and somewhere around Alpha. Even IA64 has more support around it than PPC. It might have been the whole IBM stuff. PPC was limited to AIX and OS/400 for the longest time. With a commodity server now, at a reasonable price, there should be more adoption of it.
Only items that would be x86 only would be precompiled binaries. If there is source, and gcc supports the call, then it should compile on any version. There would be x86 type calls only, especially with C++ and such, but it should work.
-
For quite some time Red Hat has been the darling on Power. They've kept a port for over a decade now, I think. Suse was always tier 1 but RH was right behind them after the first year or so. IBM has been making sure at least one of the two leaders was continuously available on everything from the little rack mount Power boxes up to their mainframes since around 2001.
-
I wonder if IBM will approach this from a hosting perspective as well. Why not build a cloud platform on top of this?
-
No reason that that would not work. Get KVM or Xen working on Power, port OpenStack and away you go.
-
I would love to see Power as a hosted product. I would be happy to try using Power for projects, but I mostly look to hosted cloud products for computing resources these days when dealing with the Intel world, why would I move back to physical to work with Power? They have cleared on hurtle but they have another big one to get over before they are on the same playing field, I think.
-
@Reid-Cooper said:
I would be happy to try using Power for projects, but I mostly look to hosted cloud products for computing resources these days when dealing with the Intel world, why would I move back to physical to work with Power?
Because PPC has always been available to "virtualize".
I work with iSeries boxes so this will be AS/400 specific, but they have a function called LPAR. If you need separate instances, all you have to do is partition the system and you have a "new" machine ready to go. This has existed in the AS/400 line for a long time, even before PPC, introduced in V4R4 back in 1998.
Yeah, it's not "virutalized" like you expect x86 to be, but it's effectively the same.
-
I'm familiar with LPARs, but will the LPAR system be available to us on the LC servers?
-
AFAIK, that's part of the whole architecture. Kind of hard to change course since these features existed within PPC well before OpenPOWER existed.
Can't seem to find the Redbook on it from IBM though, although keep in mind Rackspace is selling OpenPOWER stuff. I doubt they are buying hundreds of these things and letting the customer at it.
-
Good point. I don't particularly doubt that it is there, just don't want to count unhatched chickens, you know? HP used to sell Proliants with the built in virtualization disabled via BIOS. Not that IBM would pull that, but without knowing that the feature is there for sure, it's something to look into.
