IBM Reveals New Virtual Linux Environment

jenwren1010 writes to mention that IBM has just announced the new open beta version of their virtual Linux environment that allows users to run x86 Linux programs on POWER processor-based IBM System p servers. "Designed to reduce power, cooling and space by consolidating x86 Linux workloads on System p servers, it will eventually be released as the [rolls-off-the-tongue] 'IBM System p Application Virtual Environment (System p AVE).' With a 31.5% global revenue share during 2006, IBM hopes to build on System p UNIX success and extend firmly into the Linux marketplace. Considering there are almost 2,800 applications that already run natively on Linux on System p servers, the chances are good that it will succeed."

Comments

What's the point?

[Ed Avis]

I don't get it, aren't almost all Linux programs able to build for pretty much any architecture? The only use for emulation would be binary-only proprietary software that's built for x86 only. And even there it should be pretty trivial for the vendor to port it to POWER.

Re:What's the point?

[timeOday]

My question is, why use POWER processors at all? Why not just run a bunch of VMs on an x86 blade server?

Re:What's the point?

[Biggerveggies]

The point is that you can run a unified infrastructure with scalable LPARS for different clients on one box (think p595).

ie - the marketing term: "Power on Demand".

Re:What's the point?

[Anonymous Coward]

Obiously you've never tried to port from Linux to AIX. Sure, "Hello, World!" ports over just fine, but beyond that you're going to have issues. The gcc tools on AIX are not good. You're forced to use the AIX linker. Shared objects, static linking...good luck figuring out the differences from what you're used to on Linux and what they mean to your program. Oh, you want to throw a C++ exception from a library? Good luck.

Trust me, this software will help a lot of people get their big apps working on AIX.

AIX is an interesting platform. I have a love-hate relationship with it. The management tools are great. The OS is rock solid and the p-series hardware is fantastic. You can scale these things in several ways and its not that expensive when you look at what you're getting. I'd rather admin 50 AIX machines than 50 Linux machines. But...when it comes to using the GNU tools on AIX, its just a lot of headaches. You can get most things working...but it just takes a lot of time. I hope this software helps with that issue.

Re:What's the point?

[Tinkster]

That's one good and fair comment; the other thing is that you can have
one of those big irons running an x86 Linux that will run your "commercial
product of choice" which is certified against a specific version of Linux
w/o having to buy x86 hardware and gain expertise in using VMWare as well...

Re:What's the point?

[Kadin2048]

I think Power (according to IBM, anyway) offers a lot better performance/watt and scales better up to supercomputer-ish sizes.

And I think you can even integrate x86 blades into some of IBM's high-end systems for running Linux x86 binaries; the idea is with this new virtual environment, you wouldn't need to purchase the additional hardware.

I see this whole thing as basically a bullet point that they can use when selling POWER to a prospective client -- they can put it out there as one architecture that will run most anything. (Well, except Windows stuff.)

Re:What's the point?

[Kadin2048]

Yeah I originally thought it was a compatibility layer that would let you run x86/Linux apps on POWER/AIX, but I don't think it's quite that.

From TFA:

IBM expects ISVs that don't already have a native Linux on POWER product to be able to expand their addressable market to System p servers at minimal cost by allowing them to run their existing x86 Linux applications on these servers without having to recompile, release new media or documentation, or maintain a unique product offering for POWER technology.

So basically it's a way of taking x86/Linux binaries and running them on POWER/Linux without a recompile. (And, one assumes, if you're an end-user, without going back to the software's manufacturer and paying through the nose for a new POWER version; you can move from x86 to POWER and still use all your same apps, without buying new versions.)

Re:What's the point?

[Chanc_Gorkon]

Some good points, however the term big iron should not be applied here. Big iron would be a mainframe. Pseries machines are very powerful, however, they are very different in technology then a mainframe. They are a hybridization of technology that IBM built with POWER and of Mainframe like techn0ologies.

This is a great idea. With micro partitioning on the pSeries and automatic load balancing, us pSeries admins don't need to learn VMware to run a farm of x86 based servers. Also, while most things are running on POWER already, sometimes it's not convenient to find binaries that will run on it plus how many of us have a spare pSeries machine just for compiles?? Also, there's a metric tone of commercial apps that run on x86 Linux and not many of them that run on PPC based distros.

Re:What's the point?

[Gooner]

Well one reason it's useful is that it's possible to partition POWER servers down to tenths of a CPU so it's easy to find space to run something like p-AVE. Another is that SLES is licensed by the box rather than by CPUs or LPARs so anything that helps get more apps to run is a good thing.

I've got a 16 CPU P570 here at work and we run Linux on it exclusively due to the cost, as AIX means that you get soaked on costlier licenses. I've done my share of trying to get apps (primarily statistical programs) to work on the POWER CPU. I got R to work but there plenty of other programs that either don't have source or won't compile cleanly though part of that is almost certainly due to my GCC n00bishness so being able to run the x86 version right away is compelling.

I'm also beta-testing p-AVE right now. It works and is easy to get up and running. It's slow right now though compared to something that can run on POWER. It's interesting that this isn't an IBM product. It is from the same company who made Rosetta for Apple, namely Transitive Corp. So in one product you're going from PowerPC to Intel and the other goes Intel to POWER. It looks like IBM are going to do what Apple did and swallow the cost for end-users (or maybe make it back in Global Services consulting fees).

I seem to remember a similar technology from SCO

[Anonymous Coward]

I think it's pretty cocky of IBM to do this while the SCO case is still before the courts. It may be a case of the hand that steals not knowing what the hand that conceals is doing This time it might get a slap!

I recommend SCOX. It's a BUY.

T_A

but does it run linux?

[Anonymous Coward]

what?

Power Saving?

[hawg2k]

It's been my experience that IBM's power architecture isn't really known for being "green". Can anyone provide some expertise behind the statement that running Linux VM's on the P hardware will really save energy in heating and cooling over other concepts like a rack of 1-U rack servers, a VMWare/Xen type solution on x86 hardware, or some type of blade solution?

Re:Power Saving?

[Chanc_Gorkon]

You can buy ONE rack and it has 64 processors in it. Each of those 64 processors can be sheared down to 1/10th of a CPU partition. That would be 640 servers in one rack. Granted, you'd never want to run that many LPARS on a system, but you could come close. You can also share all 64 of these processors and each of the LPARS can look like a SMP system by setting a VP of 2 or higher. Granted alot of this will need proper tuning, but you can do a lot with a pSeries and shove alot of hardware into one rack. Also, with Partition Load manager, it can very how much CPU each partition gets by the load it's getting. Say one of the LPARS gets nailed all of a sudden. If the partition is uncapped or not reached it's cap, it can automatically grab as much CPU as needed.

The pSeries machine CAN do what they describe.

Re:'nix on p Series? Run AIX.

[cgh4be]

I think there are a couple of reasons:

1) There are some (not many, but some) applications that run on Linux that don't run on AIX (i.e. won't compile on AIX)
2) There are a lot of Linux gearheads out there that a company might not want to retrain for AIX

The whole point is to be able to run (almost) any operating system you own on (almost) any platform IBM sells. If Windows and Intel weren't in bed, Windows would be running on the pSeries. In fact, it is in the lab, it's just not for sale :)

This is the point.

[HockeyPuck]

In the current generation of Power CPUs, you can implement micropartitions, akin to "this partition uses .1 CPUs", which if you've got spare computational power available on your AIX system, you could create additional partitions for X86 use. Also, since the partitions have the ability to communicate directly with each other without going over an external network, you could have in one chassis an AIX database with a linux based webserver in different partitions, both sharing the same fibrechannel cards and external gigE/10Gig network connections.

More details

[aktzin]

The article and press release don't say much, but I found this announcement on the IBM web site: http://www-03.ibm.com/systems/p/linux/systempave.h tml

At the bottom are some good details:

"Runs most x86 Linux applications except those that * Directly access HW; * Are hardware architecture specific; * Provide unique kernel modules; or * Use instructions added later than the Pentium II processor, e.g. SSE2."

"All application components and plug-ins must meet these qualifications. Support for x86 Linux applications requires an Red Hat 4 update 4 or Novell SLES 9 with Service Pack 3 of the Linux operating system."

Re:More details

[Lorkki]

This likely isn't something that would just work on your PowerPC desktop anyway. The POWER architecture is aimed at a completely different market, for bigger servers and the like, and the instruction set is a superset of PPC's. Moreover, games these days use various HAL APIs to talk with acceleration hardware rather than directly accessing it - you'd want something like WINE with an x86 emulator, but DOSBox is available for PPC platforms if you need to run those older titles.

Finally PPC used for what it was designed!

[PaulBu]

As far as I remember one of the original goals of PPC architecture in the times of original IBM/Moto/Apple consortium 15 years ago was to be able to emulate "other" (x86, maybe? ;-) ) processors efficiently. Strangely I have not heard about something like this being actually used up until today! (Yes, I know that POWER != PPC, but I think the parts are still there).

Paul B.

Re:Remember folks...

[Constantine XVI]

umm, AT&T? Standard Oil? The Holy Roman Catholic Church?(sorta j/k on the last one)

Re:Imagine an array of virtual Linux machines

[Constantine XVI]

And then what, set them up in a Beowulf cluster?

Never would have thought of it before.

Re:Imagine an array of virtual Linux machines

[WillAffleckUW]

Why yes, a virtual Beowulf cluster!

But, first, we could argue over whether they would work better with BSD or Linux ...

Re:Too late

[Lemming Mark]

Qemu could run an x86 OS inside a PPC OS. Actually, Qemu can provide a user-level binary translation layer to apps, including translating syscalls appropriately - you don't have to emulate a whole system, the app has its own sandbox that looks like the foreign architecture.

How I think they'd answer:

[Kadin2048]

I can buy a lot of commodity hardware and power it for $100,000. Let's just say a decent Pseries will be an order of magnitude more expensive for the initial purchase, never mind the annual support

True. In my (admittedly limited) experience though, IBM hardware generally gets aimed at organizations whose IT budgets are already fairly big (I won't say "bloated"), and are paying through the nose for support already.

If you're looking at commodity servers and supporting them yourself, you're probably not going to look at IBM; their customers are going to be choosing between IBM pSeries, and maybe Sun's high-end SPARC gear, or maybe HP 9000 series stuff. They're probably migrating up from superminis with atrocious support costs anyway (and they may only be migrating because their superminis are being EOLed -- I've run into lots of organizations who were perfectly okay paying the support for their legacy gear, until it was no longer supported), so a $100k IBM system could easily look like a savings over 5 years when you consolidate a dozen "small iron" Unix boxes onto it.

I'm not exactly sure how they would find a cost savings if you were already just using cheap x86 servers, though. I guess they'd probably say 'consolidation,' but I don't know exactly how many commodity pizza-boxes you'd need to consolidate to pay for the TCO on a pSeries... I guarantee though if you called an IBM sales rep, they'd be able to make the numbers work, somehow.

IBM's own page on "Why Linux on the POWER?" is fairly interesting:

The IBM System p(TM) server family and the IBM BladeCenter® JS21 blade server are packed with features designed to enable you to achieve lower costs and more flexibility, as well as have the peace of mind that comes from knowing your applications are available when you need them. Our leadership performance saves you money by providing exceptional performance per processor core and including up to 4 cores per socket. Unique IBM virtualization technologies are designed to dramatically increase server utilization by providing innovative capabilities that enable one server to act like many--while giving you the ability to automatically move more processing power to critical applications when needed. You can meet known and unknown processing requirements with fewer servers -- so hardware, software and facility costs go down. Finally, your Linux® applications on these systems will be available when you need them thanks to time-tested IBM reliability features.

I think they're going for PHB appeal here. The idea is that you have one machine, one support contract, to one company, and that's the end of that. (In theory.)