“Today, IBM announced a public beta trial of a virtual Linux environment that will let x86 applications run on its System p Unix servers without modification. The new IBM System p Application Virtual Environment technology will allow x86 binaries to run as well without modification, removing the biggest barrier against effective virtualization for some companies. As a result, customers will be able to consolidate dozens, if not hundreds, of servers into one virtual environment.”
but if running Linux apps is desired, why not just buy IBM System p servers with Linux pre-installed rather than a proprietary Unix?
but if running Linux apps is desired, why not just buy IBM System p servers with Linux pre-installed rather than a proprietary Unix?
Because then IBM can’t sell you AIX, right? ;-p
Because then you can’t run your AIX apps. This handy if you have mainly AIX apps, but also want to run a couple of linux x86 apps.
I think you’re missing the point here. This feature is for running x86 Linux binaries under POWER Linux. So, yes, you’d be running System p and Linux, but running x86 binaries on it.
chicklin,
Really? Heh.
But I thought you could just recompile most software for the PowerPC/POWER architecture and have it work natively.
Linux software that uses low-level x86 ASM could just be ported to POWER ASM?
“but if running Linux apps is desired, why not just buy IBM System p servers with Linux pre-installed rather than a proprietary Unix?”
For this feature (pAVE) you’re not running a proprietary UNIX (AIX), you’re running Linux on POWER but running x86 binaries.
Yes, most FOSS applications probably can just be recompiled, but there are hundreds (if not thousands) of commercial applications out there that are currently x86-only and you don’t get the source code. Those apps are what this feature is targeted at.
I love FOSS as much as the next guy, but in the business world, especially large enterprises where this kind of hardware is typically purchased, FOSS doesn’t always cut it.
The important thing here is that system p are Powerpc based (Power 5) and this technology will enable x86 binaries to run on it without modification.Is like Rosetta for MacOS X but in reverse
Sounds very similar to lxrun on Solaris..
http://www.sun.com/software/linux/compatibility/lxrun/index.xml
The problem with these implementations is that there really isn’t any 3rd party support for them. Nor any real demand.. Afterall, why run the Linux x86 binary for (insert application name here) when you can just run the native version? Certainly they’re not suggesting you run some commercial x86 Linux software for which there is no native version under this mechanism.. right? I highly doubt that would be a supported configuration under the support terms of those 3rd party providers.
So I’d be interested to see an example use case for this. And not including OSS for which I could simply compile native versions nor any proprietary software for which my support would become void.
I think that’s exactly what they’re suggesting. ISV’s would still need to certify it, but that would typically be a lot easier than porting the code.
Sounds very similar to lxrun on Solaris..
Not exactly. According to the FAQ, lxrun only works on x86 Solaris, not SPARC. This tool seems more (as someone pointed out) like Rosetta for Linux x86 -> Linux POWER.
So I’d be interested to see an example use case for this. And not including OSS for which I could simply compile native versions nor any proprietary software for which my support would become void.
While the tech is neat, you’re right, there’s not a wide need for this software. It’s not something I’d build my large parts of my infrastructure on. But, if I had lots of POWER servers, and some piece of software that was critical and x86-only, it would be useful.
Not exactly. According to the FAQ, lxrun only works on x86 Solaris, not SPARC. This tool seems more (as someone pointed out) like Rosetta for Linux x86 -> Linux POWER.
Yea, you’re right. lxrun / Sun’s project Janus was the first thing that came to mind when I read the article and although they are not exactly comparible technologies they share many of the same issues that will prevent adoption.
The ISV’s go through a lot of work to develop, test and ultimately support their products on the major Linux distro/architecture combinations, namely RHEL and SLES on x86. So unless you can run a full x86 RHEL or SLES install under pAVE and it is supported by Red Hat and/or Novell I dont really see this going anywhere.
But, if I had lots of POWER servers, and some piece of software that was critical and x86-only, it would be useful.
If it was critical, would you really want to run it in this sort of unique config or on the tried and true (and likely far better supported from the app vendor) x86 configuration?
The article states:
Although approximately 2,800 Linux-based applications run on IBM’s POWER processor-based System p servers, there are tens of thousands which will run only on less advanced x86 servers.
I guess the POWER guys forgot to actually look at what the x86 guys were doing. IBM sells highly scalable x86 servers today. The x3950 scales up to 32 processors (64 cores) and 512GB of memory..
http://www-03.ibm.com/systems/x/scalable/x3950/index.html
If I were looking to consolidate my x86 Linux workload on an IBM platform, I’d be looking at IBM’s X Architecture. The x3950 running RHEL or SLES with Xen (or even VMWare ESX) would be a very nice consolidation platform AND would not require any funky emulation or syscall mapping. It has the additional benefit in that you could not only consolidate your Linux workload, but your Windows workload as well.
Yes, this is a Linux on x86 to Linux on POWER translation layer, not to be confused with a Linux on x86 to Solaris on x86 translator.
Linux on POWER is provided by both Red Hat and Novell. While ISVs that already certify for RHEL on x86 and/or RHEL on POWER might have to re-certify for RHEL on POWER via pAVE, no new test cases or procedures would need to be developed. It will cost them time and money, but not a whole lot.
While pAVE might be new and not exhaustively tested, it’s quite likely that pAVE could offer more reliability, availability, and serviceability than could be achieved with Linux on System X. Scalability isn’t the only advantage of System P hardware. P is essentially a (slightly) more affordable mainframe. There’s a dedicated service processor for the firmware, non-volatile memory for persistent data storage, advanced virtualization support baked right into the hardware, tighter oversight of bus devices, a hardware management console, and much more. This is serious hardware, while even System X is just a commodity Intel system on steroids.
In the end, IBM is a hardware company. pAVE is a way to sell more System P at a time where P is years ahead of the competition in the high-end UNIX space. They want you to bring your commodity software to their high-end hardware, because they really don’t care that much what you run as long as you buy the box. All of IBM’s systems products support Linux and at least one other software platform.
As an AIX developer, I look on with curiosity. Linux will gradually sunset AIX, but how soon, and how quickly? When I talk to my colleagues about Linux and how fast it is catching up to the high-end server platforms like AIX, the question they often end up asking me (or themselves) is, what does AIX have than Linux doesn’t? A whole lot of complexity and questionable features, but not that much else. The part of AIX I work on didn’t even exist in Linux a year ago, but now it has a rapidly advancing implementation with a design that’s arguably superior to the way we do it in AIX.
The time for AIX to begin its ride off into the sunset seems to be approaching, and pAVE seems to be a step necessary to move Linux into a better position to be a replacement platform for AIX users down the line. However, AIX isn’t going anywhere just yet. AIX 5.4 will have somewhere in the area of 1.5 million lines of new code in the base operating system.
Edited 2007-04-25 01:13
not to be confused with a Linux on x86 to Solaris on x86 translator
I’m not confused with differences, rather what I was attempting to highlight were the similarities in the issues with adoption between the two technologies.
Scalability isn’t the only advantage of System P hardware
I’m intimately familar with running Linux on System x, System p and System z as well as the merits of each hardware platform. And there are certainly use cases that are compelling for each. But as time goes on the non-x86 solutions are becoming less and less compelling.
no new test cases or procedures would need to be developed.
That is not a true statement. IBM states on the qualification page (http://www-03.ibm.com/systems/p/linux/qual.html) software that directly access hardware, take advavtage of non-POWER kernel modules, use instructions beyond P2 or make direct access to physical memory are not supported and therefore not likely to work.. So clearly “new test cases” and/or “procedures” would need to be developed for a lot of software.
The time for AIX to begin its ride off into the sunset seems to be approaching
I’d argue it has already started and not just for AIX, but for Unix in general. It certainly has at the company I currently work for and for companies I have consulted to previously. I agree that IBM in the end wants to sell you the hardware which is exactly why they are pushing Linux on POWER and on System z. And if they can convince some people to go that route and prolong those platforms a bit more, than great! But I think it’s a bit foolish to think that in the end x86 is not going to be the one left standing as the premier Linux architecure. So what will that mean for those traditional Unix architectures once those Unix OS’ have disappeared? I’m comforted by the fact that they are, although perhaps quitely, doing serious innovation with x86 as well. And as someone who has experience with Linux on nearly all IBM’s platforms I can tell you that it works best and you get the best support experience when it is running on their x86 systems.
This is serious hardware, while even System X is just a commodity Intel system on steroids.
While I agree that System p has some features not available in the X3 architecture product (yet), calling it “a commodity Intel system on steroids” is a gross misreprestation. X3 is anything but a commodity system and has in fact quite a few features borrowed straight from System p/i/z and others that are at the very least complementary. The only significant thing X3 has in common with commodity Intel systems is the CPU, so I’d say that’s a bit more than just steroids going on. And having been recently exposed to some insight as to what is coming with X4 and X5 I see these differences only getting smaller.
But as time goes on the non-x86 solutions are becoming less and less compelling.
Or rather, x86 is getting more compelling. 64-bit addressing, VT-x, SVM, nested/shadow page tables, and more are bringing x86 closer to other architectures than it was before.
So clearly “new test cases” and/or “procedures” would need to be developed for a lot of software.
No, unsupported software would just fail the same testcases it passed on a native x86 system when run with pAVE. Software that trips up the translator isn’t going to get very far in testing at all. The code still does the same stuff and has the same execution paths. It’s just running in a different environment. You have to rerun the tests, but you don’t need new tests.
But I think it’s a bit foolish to think that in the end x86 is not going to be the one left standing as the premier Linux architecure.
Why? All of the gaming consoles are PPC-based. x86 hasn’t fared all that well in the ultra-portable market where much of the growth will be over the next decade. Linux will always work best on the architecture with the most users, but it works well enough on other architectures that there really isn’t much tying Linux users to x86. If another architecture becomes more popular, Linux will begin to support it better. I’m not arguing that this will happen, but that it could happen, and that there is nothing special about x86 that makes it the best architecture for Linux.
I agree that System p has some features not available in the X3 architecture product (yet)…
Yes, X, P, and Z are essentially racing to provide a similar feature set to different sectors of the market. Everybody wants density, efficiency, utilization, manageability, and reliability. That used to mean completely different things to these markets, but now its the same features scaled to the appropriate level. In the BladeCenter environment, we have the ability to put a bunch of X and P blades (and now storage blades) into a single box and manage them through a single interface. P and Z, along with I, are standardizing on a single platform architecture based on POWER6. There is convergence, but in order to scale from the SMB to the retail chain to the financial institution, we need different sized boxes with different sized approaches to these common demands.
Linux on P is a different sized approach to Linux than we offer on X. Both are premium systems with premium features, but they’re for different kinds of customers.
First off, good discussion, thank you..
You have to rerun the tests, but you don’t need new tests.
And if it fails and you actually care and perhaps want it to work at least at some level then you’ll need to go back to the code make the modifications and test again. Perhaps even compromising on certain functionality. If you can get away without ever changing your tests and procedures during this time and those tests/procedures are valid, then great!
All of the gaming consoles are PPC-based.
Yes PPC based chips are common in these devices and the embedded market for obvious reasons. There are lots of different classes of PPC chips and many will have a market and life span for many years to come. Look how long the Motorola 680×0 lasted in the embedded market well after the last of those chips were making it into servers and desktops. My opinion is that x86 or compatible chips (maybe time to dust off the old PPC 615 designs) since as you put it are “getting more compelling” and already have the most momentum are the ones to bank on in the server and desktop use cases. Especially now as AIX based server sales have leveled off and by most accounts will eventually begin to decline and Apple (the biggest PPC desktop integrator) no longer produces PPC based desktops. So x86 capabilities are increasing, their server/desktop deployment rate continues to climb and costs are coming down. While the PPC deployment rate in server/desktop are decreasing and costs remain high..
there really isn’t much tying Linux users to x86
Well just the obvious stuff like availability of software, cost of x86 hardware, general availability of x86 hardware and steadily growing momentum that goes back more than a decade…
There is convergence, but in order to scale from the SMB to the retail chain to the financial institution, we need different sized boxes with different sized approaches to these common demands.
Which is exactly how IBM markets the x3950. As your needs grow, simply slap in a new quad (up to 8 total). So you can go from 4U (4 sockets) to 32U (32 sockets) all without leaving the comfort of x86.. And the newer X architecture boxes in the pipeline are even more scalable and have even more RAS capabilities. Now I’m not saying this is a panacea, but it does meet the needs of a significant portion of business from small to large while reducing overall complexity and costs and providing the benefit of true x86 compatibility and the associated software availability.
Linux on P is a different sized approach to Linux than we offer on X. Both are premium systems with premium features, but they’re for different kinds of customers.
As I pointed out with the x3950, the p and x products are not mutually exclusive. X architecture (X3/4/5) is rapidly closing the gap and already meets the needs for a large number of enterprises that would have chosen p for the very same workload just a few years back.
IBM, by there own admission is a services company. How pAVE fits in with this overall strategy baffles me. Most likely it is a play to bridge the functionality gap between X86 and POWER Linux and little else more.
People have been saying this for years. Linux is ‘on the cusp’ of being an enterprise player. UNIX is on its last legs. Sorry to be blunt, but its all BS. People make these claims as if UNIX technology is standing still. System P is currently the finest virtulization plaptform available today. It makes VMware infrastructure look like a joke, and thats coming from a VCP.
Before people are so quick to put a nail into proprietary UNIX platforms ask yourself:
What virtulized platforms do Oracle support?
What virtualized platforms do SAP support?
How many processors can a x86 virtualization platform support in a guest? Compare and contrast that to System P.
Can you reallocate (hot) memory, processors, IO adapters on X86 virtulization platforms?
I love Linux for a lot of things, but I’ve seen waaay too many pushes for Linux to run enterprise caliber applications, only to fail and have to be redeployed to UNIX. Ive got my own war wounds in this reguard in my more idealistic days.
Take care -c-
Heh…
Even if there was complexity involved in porting x86 apps, just guess how many distro repositories there are.
IBM could just grab the source code from any PowerPC/POWER repository and compile it to work on their System p Linux servers.
IBM can even host their own private software repository for customers.
This feature is targeted at x86-only, binary-only Linux apps where you don’t have the option to recompile. Not every piece of software comes with the source code.