Slashdot Mirror

← Back to Stories (view on slashdot.org)

Dynamic Logical Partitioning for Linux on POWER

Posted by Hemos on from the learn-how-to-do-more dept.

An anonymous reader writes "Logical partitioning provides POWER processor-based servers with the capability to do server consolidation and optimize system resources. Dynamic logical partitioning enhances this capability by providing control of the allocation of the resources without impacting the logical partitions availability. Linux on POWER supports dynamic LPAR for changes to physical I/O, virtual I/O, and processor resources."

111 comments

  1. droool... by dj245 · · Score: 3, Funny

    And for once I drool over something I have only vaguely an idea of what it does.

    --
    Even those who arrange and design shrubberies are under considerable economic stress at this period in history.
    1. Re:droool... by Amiga+Trombone · · Score: 5, Informative

      And for once I drool over something I have only vaguely an idea of what it does.

      What it does is allow reconfiguration of system resources, such as IO cards, memory or cpu's (or on Power 5 with AIX 5.3, portions of a cpu), etc. on the fly without having to reboot your server to acknowledge them. AIX has had this capability since 5.2.

      It's great for being able to juggle your resources on the fly, but it really comes in handy for moving your DVD drive between partitions on a frame without having to reboot. Having to reboot 2 servers just for that is a royal PITA.

    2. Re:droool... by frodo+from+middle+ea · · Score: 1
      can you also care to explain , why can't this be done on our beloved i386 line ?

      has it got to do something with the CPU ? what about memory and I/O cards ?

      --
      for the last time people, I am "frodo from middle eaRTH", not "middle eaST".
    3. Re:droool... by Ryan+Amos · · Score: 3, Informative

      The POWER series has this on-chip, so it's a whole lot faster than doing it via software and it doesn't require a reboot. The dynamic partitioning is the real difference between the POWER series and the PowerPC chips IBM sold to Apple. This is a feature carried over from IBM's mainframe days, and if you actually need it, it is very cool.

    4. Re:droool... by sapbasisnerd · · Score: 1

      You need a few things to be able to do this. One is what IBM calls the hypervisor, this is essentially privledged instructions on the chip that are used to control access to underlying system resources. These instructuions are isolated so that the operating system cannot execute them. Two is a service processor, this is a separate, separately powered special purpose processor that CAN access the hypervisor instructions it also does things like control power etc.. Additionally you need to give each PCI card it's own bus so that a card cannot issue bus level events that will disrupt other cards. You could build a x86 server with a service processor (in fact IBM already does) and bus per slot PCI and still not be able to do this beacuse Intel hasn't built the virtualization instructions into the x86 architecture.

    5. Re:droool... by Anonymous Coward · · Score: 0

      on-chip isn't fair

      it's part of the overall design, but it's not on the CPU

      on the AS/400^WiSeries^Wi5, this is all controlled by a standalone management system (the HMC, which i hear came over from the AIX side). the HMC talks to the service processor in the i5. the service processor talks to the hardware and does the reconfiguration. the service processor is a computer in and of itself (even runs a web server!)

      actually its a lot more complicated than that even. partitioning has to be planned extensively. some things in the partition are fixed (load source, the IO controller for the load source) and some things are dynamic (memory, additional IO controllers, etc). the coolest part about partitioning is that you get to talk between different partitions in the same hardware using virtual ethernet.. that's a fast network!

      anyway, I personally love these articles on slashdot. the majority of readers here run linux on x86 clones and these articles are about some of IBMs over-engineered servers (note that over-engineered is not meant as a jab but as a compliment)

    6. Re:droool... by rshimizu12 · · Score: 1

      DLPAR is way cool.................. DLPAR is generations ahead of any VM technology on the market. The big difference is that DLPAR utilizes a hardware based hypervisor that manages the different partitions. DLPAR has the ability to provision or de-provision new lpars on the fly based upon a servers load. LPAR's go way back with IBM's Z-Series........

  2. Cool, but still buggy by Kris2k · · Score: 4, Interesting

    I've worked on it on a few times, and its still a bit buggy, but IBM seems to never cease to amaze me by pulling-out new patches on a daily/weekly basis. With time, this technology will perfect itself, and when it will, it will really rock, for now, I'd still go with a BladeCenter + SAN.

    1. Re:Cool, but still buggy by Kris2k · · Score: 2, Interesting

      Oh, and an after-thought:

      Redundancy is achieved with multiple independent subsystems

      Well, in the case of LPAR, when your host OS decides to pull a quickie on you, you find yourself on a standstil with an entire company of 100+ users without their apps running. I wouldn't like to be in the shoes of the iSeries admin in that case.

    2. Re:Cool, but still buggy by rhaig · · Score: 1

      blades don't buy you much. they still price about the same as a bunch of 1U boxes, and if you want hot-swap scsi drives, bladecenter doesn't increase density. if you go with boot from san like you're implying, your density doubles, but I've not known anyone who's had a good experience with boot from san. You can go with the IDE drives and still have the double density, but do you really want ide in your production environment?

      --
      "We are not tolerant people. We prefer drastically effective solutions"
    3. Re:Cool, but still buggy by atomicdoggy · · Score: 2, Interesting

      An LPAR doesn't have a "host" operating system. The partitioning is done in hardware (each LPAR has it's own installation of the operating system).

    4. Re:Cool, but still buggy by Kris2k · · Score: 1

      It depends.. (do note: I'm not the expert, I just was forced to play(hack) with it when things didn't work)

      The OS/400-centric implementation would have a tendancy of managing the partitions from the OS400 operating system, feeding-off virtual scsi devices to the neighbour partitions, which is logical in a way, since you confine all your backups to one operating system.

      The LPAR will pool your cpu and memory ressources, but you still have to feed it an I/O subsystem for Disks, Ethernet, etc.. I aven't seen it do that, except assign physical ethernet devices. Which means that you can't share a CDROM, a Disk, an Ethernet card, unless you have a partition managing this, in this case, the OS/400, maybe there is a way, but I'm not the expert, I just implement the guest Linux partitions.

    5. Re:Cool, but still buggy by ag3n7 · · Score: 1

      Well, yes and no. You can configure the LPAR to be entirely segregated from a hardware standpoint. However, for the best rate of consolidation, you'd need to use VIO servers (virtual I/O) for the sharing of disk (at an Logical Volume level) and the sharing of Ethernet cards (since you can't split the 2 port cards over 2 LPARs). In that case, the "loss" of the VIO server would impact any client LPARS.

      Which is why a lot of people create two VIO servers, to provide redundancy across two seperate LPARS. That way, if you lose one VIO server, your client LPARs are not necessarily down.

      The VIO servers are basically "appliance LPARs" in any case. They have no other roles except to provide "I/O Servers" to client LPARs.

    6. Re:Cool, but still buggy by vdthemyk · · Score: 1

      Well, the interesting thing is, with the i5 boxes, they dumped the manditory host partition for linux or AIX partitions. On the p5 systems, you can still have dedicated resource partitions (own disk adapters and such), but they also have the option of running a host partition for shared resources like ethernet or disk. It's called the VIOS. It's basically a skinnied down version of AIX and all it does is manage the physical resources that are being virtualized to the client partitions.

      --
      VD
    7. Re:Cool, but still buggy by atomicdoggy · · Score: 1

      That is correct, but, DLPAR (Dynamic LPAR changes) is not the same thing as VIO (they are separate concepts/features).

    8. Re:Cool, but still buggy by janic · · Score: 1

      To expand on the other reply you got:

      Ther are two ways you can do LPARs. If you have enough available, Physical IO resources for each LPAR, then you can assign each LPAR it's own NIC, Hard Disc controller, SCSI controller for Tape, etc.

      If you don't have enough Physical IO, then you can have an instance of your favorite os (Linux, AIX or os/400) "own" all of the physical IO and serve out it's resources to the other LPARS.

      _or_ you can have a mix of the two. For example, you could have one LPAR running Linux for iptables, squid, etc. with it's own NIC as a firewall, another LPAR running, say, AIX, DB2 and (shudder) websphere as an app server, and a third LPAR doing file/print etc. with it's own nic. Each partition owns it's own disks (RAID controllers, actually). Now, through the HMC, you can add a virtual nic to each partition, (which appears as a physical device to the OS inside the partition,) configure them appropriately, and you have a DMZ network, and the equivalent of a small server room in a single box.

      As far as architecture goes, the power5 boxes are NUMA. The (multi-core, yay!) CPUs each have their own RAM, but will use ram belonging to other CPUs if needed. The service processor plays traffic cop and only allows a partition to access resources that have been assigned to it, as well as being the physical embodiment of all of the virtual devices.

      So... in reality, the Service Processor is about the closest thing to a host OS on these machines, that is, if you are not using any VIO. It's pretty robust, and is even hot swappable on some of the higher end boxes.

      That said, if it crashes, ALL of your LPARS come to a grinding halt.

      Cheers.
      John

    9. Re:Cool, but still buggy by afidel · · Score: 1

      After walking into a number of clients who a previous consultant had sold HP BL10e's to I can catagorically state that as a big fat negative. Those things had their laptop HDD's die left and right after just 2-3 years, much less reliable that just about any storage solution I have ever run into other than the troubled IBM Deskstar line.

      --
      There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
    10. Re:Cool, but still buggy by Anonymous Coward · · Score: 0

      At least the IBM blades come with internal SCSI these days so no you do not need to run with IDE drives to have high density blades.

    11. Re:Cool, but still buggy by sapbasisnerd · · Score: 1

      That's the way it worked with POWER4 and before. POWER5 and i5OS use the microcode hypervisor and there is no host O/S anymore.

    12. Re:Cool, but still buggy by sapbasisnerd · · Score: 1
      You can go with the IDE drives

      The current 2 Way (HS20) Xeon DP EM64T based blades use SCSI drives (not hot swapable). So you can put 14 2 way servers in a 7U chassis. In addition to just density this also helps on saving other infrastructure for network cabling, KVM and power.

      The 4 way blades (HS40) are still junk, old IA-32 only Xeon MP chips, IDE drives etc. If you have an application that must scale above 2 Way then don't use a blade for that application.

    13. Re:Cool, but still buggy by JhohannaVH · · Score: 1

      The problem is, they can't get the BladeCenters stable. :( I can tell you we did 4 BIOS upgrades just in the first half of 2005 on 4 fully populated BladeCenters. That SUCKED! All cuz they couldn't get the BIOS right, causing all traffic to just stop. Hrm. But they got it where it's at. My new company wants to swear off of them, but they can't STAND the HP or Dell blades.

      Jho

      --
      Sorry man... the Internet pooped on me.
  3. Power Rocks by as400tek · · Score: 2, Interesting

    Then entire world will move to POWER in the next 10 years. POWER 5 is where it's going folks. Great IBM Hardware is paving the way for the great OSs of the world to run like champs. I have been an AS/400 now iSeries Admin for over 15 years and POWER/5 is awesome. Good to see some Slashdot coverage on the topic of POWER. IBM is still trying to figure out what to do with LINUX and maybe this is it. Will have to wait and see what happens next.

    --
    David Vasta iSeries(AS/400) Admin & Junkie
    1. Re:Power Rocks by Anonymous Coward · · Score: 0

      POWER gear is too expensive, if IBM want market share they're going to need to offer SERIOUS subsidies to developers.

    2. Re:Power Rocks by imsabbel · · Score: 4, Insightful

      They told the same thing 10 years ago about powerpc, and look what has happened.

      Target market for Power(5): Servers. No mainframes (as those are a different area), no HPC (horrible FPU/$ performance compared to main competitors), no small servers (PPC970 is more or less dead in the water, and power5 with its horribly expensive MCMs isnt cost effective in the more "normal" enviroment.

      SO i _seriously_ doubt "the entire world" will be using power in 10 years. They can be happy if the keep their market share.

      --
      HI O WISE PRINCE. WHT TOOK U SO DAM LONG?
    3. Re:Power Rocks by fiber0pti · · Score: 1

      This news is great to see on Slashdot. I think companies will start to adopt the POWER architecture mainly due to it's long term roadmap. POWER 6 and 7 are going to be even better. The only thing that will slow adoption is price. They are quite expensive which was one of Apples choices to move to Intel.

    4. Re:Power Rocks by $RANDOMLUSER · · Score: 4, Funny
      > Then entire world will move to POWER in the next 10 years.

      I respectfully and completely disagree. The world enjoys using a 64-bit extension to the 4004 architecture. We like using a single-accumulator processor with 3 "general purpose" registers. We adore the massively irregular instruction set, we like saying "push bp/mov bp,sp" every four instructions. We like the whole notion of putting values in certain (and only those) registers, so we can say "repne scasb", or "mul" or "div". The segmented memory architecture and the segment registers, are, in a word, brilliant. The notion of "near" and "far" calls and jumps, and the fact that the segment and offset are pushed in the wrong order is an endless source of delight for us. The floating point unit, and its instruction set, are nothing short of poetry in silicon. The pipelining and branch prediction are the the epitome of efficiency.

      In other words, you are just another sadly mistaken fanboy of an inferior processor architecture.

      --
      No folly is more costly than the folly of intolerant idealism. - Winston Churchill
    5. Re:Power Rocks by DannyKumamoto · · Score: 2, Interesting

      "SO i _seriously_ doubt "the entire world" will be using power in 10 years. They can be happy if the keep their market share."

      With PS3, Xbox 360 and Revolution containing some flavor of POWER5 (Cell certainly has inherited POWER5 technologies), I'm pretty sure "the entire world" would have these powerful capabilities. Unlike workstations and PCs, these game machines won't be so hacker friendly but then I doubt if most people of "the entire world" would care less....

      --
      Danny Kumamoto
    6. Re:Power Rocks by vdthemyk · · Score: 1

      I wouldn't say POWER is the way everyone is going. I would instead say "virtualization" is the way everyone is going. While POWER architecture and IBM has a big lead in this, Microsoft, HP-UX, Sun Solaris, Intel, and AMD are making headway. With the success of VMWare, hardware and OS vendors are looking for ways to capitolize on this trend of system utilization.

      --
      VD
    7. Re:Power Rocks by brunson · · Score: 1

      I have yet to see anything other than a statement that AMD and Intel will be adding features to their next line if chips that will allow Xen to run virtualized OSes without modification. But, if that comes to pass, we may be looking at the gap between Power/5 and AMD/Intel virtualization becoming much smaller.

      --
      09F911029D74E35BD84156C5635688C0
      Jesus loves you, I think you suck
    8. Re:Power Rocks by as400tek · · Score: 1

      Well then in that case the next SONY Playstation would have be a failure for your comments to be true, and since I would guess that I am going to get a new Playstation and so is the 10 guys near my desk I can assume that the majority of the world will be using Power/5 and the same technology that makes the iSeries work like a champ. Althought on a smaller scale but none the less not to far from what is running my iSeries right now, just in a smaller form factor. Now you can take the Power/5 hardware that runs the iSeries and run Linux, AiX, OS/400 or i5OS, zOS is next and the list is growing. IBM has a few large deals in the works like SONY to keep the Power/5 moving out the door. So have no fear the Power/5 is the next big thing and the way IBM is positioning it there should be many more things in our world that are running on Power/5..... Next thing you'll tell me is that Playstations and Blenders do not count as real devices?

      --
      David Vasta iSeries(AS/400) Admin & Junkie
    9. Re:Power Rocks by as400tek · · Score: 1

      I would agree with that. Virtulization is a very hot topic right now.

      --
      David Vasta iSeries(AS/400) Admin & Junkie
    10. Re:Power Rocks by as400tek · · Score: 1

      It's really not mush more expensive.

      If you are looking for a $999 server I think you might be looking in the wrong place. If you are going to haul freight you don't go get a Toyota Pickup with a 4 banger. You go get a Big Rig and get hauling. If you have an application that is a heavy load you go get a Power/5 and you might pay more but you know it is going to hold up under the strain of the load. You can get a small Power5 System from IBM for around $9000 and the same HP or DELL would not cost much less, but then again you need to think about downtime and server reliability.

      I can get 99.999% uptime out of my iSeries that was a "Cheap" one, We don't have any "Cheap" windows or Novell servers here that can stand up to that and you have to consider too the IBM support network and hardware replacement team too. I can get stuff replaced in hours not days or weeks. IBM shows up and replaces the failing part....for a cost, but it is well worth it for a 24/7 data center. In the end the Power/5 package is far cheaper long term than and Intel solution. No matter what OS you are running. Although the only OS you can run on Power/5 are so freaking stable you actually get to be an admin vs. be a patch boy all the time with your "Cheap" server running Windows or even Linux in some cases.

      --
      David Vasta iSeries(AS/400) Admin & Junkie
    11. Re:Power Rocks by as400tek · · Score: 1

      What did you say....huh....Fanboy? Yes. You have to be a fan of something! I have no idea what you were saying, but I sure wished I did. Man you sure are smart. Sadly mistaken. I really don't think so, but time will tell and so far what we have seen is I am winning that argument already.

      --
      David Vasta iSeries(AS/400) Admin & Junkie
    12. Re:Power Rocks by as400tek · · Score: 1

      I like you Brunson (Not like that.) your my kind or "Sailor" and Bad-ass and latley I am not sure if even Jesus loves me. Thanks for the comments. Nice to see that there is someone smart in Colorado. Rock On Power/5....I need to keep my job!

      --
      David Vasta iSeries(AS/400) Admin & Junkie
    13. Re:Power Rocks by $RANDOMLUSER · · Score: 1
      > Man you sure are smart

      No, just scarred. I love the Power architecture, and I'm making fun of the x86. It's like shooting fish in a barrel, and I could have gone on. I do hope your prediction turns out to be true.

      --
      No folly is more costly than the folly of intolerant idealism. - Winston Churchill
    14. Re:Power Rocks by saider · · Score: 1

      If your post were an advertisement, the following legalese would be stated near the bottom.

      "10 guys near my desk" is not an accepted marketing survey technique and the results should not be used to estimate global demand. Other demographic groups, particularly females, may show significantly different preferences. Sony and IBM are registered trademarks. So are Dell, HP, Gateway, Intel and AMD but since they are not using POWER this statement is not needed.

      --


      Remember, You are unique...just like everyone else.
    15. Re:Power Rocks by arobar · · Score: 1

      Since when are mainframes are a totally different area? The IBM zSeries machines now use the POWER5 chips, and have even better LPAR functionality that is mentioned in the article (more partitions, more processors per system, etc).

      Also, no small servers? I don't know who you've been buying from, but tear up your contract with them right now. IBM offers really low end SMB servers (ie. the i5 520 Express Edition) that use POWER5 chips and have the same pricepoint as their Intel-based xSeries brothers.

      I wouldn't predict a global shift to POWER chips within 10 years, but certainly the entire server market could easily be dominated within that time. IBM's Linux oriented push over the last few years has really strengthened the POWER5 offerings. Allowing a free OS onto hardware with these LPAR and micro-management capabilities, with price points as low as they go could easily push these servers into the office of every Joe Server Admin who wants to make his management duties a little easier (not to mention his server performance a little better).

      While writing a book on the capabilities of the iSeries systems, I used one of the medium sized machines for 10 hours everyday, 6 days a week for 2 months solid. Being able to run 3 independent Linux OSes and a the i5/OS simutaneously was incredible. Dynamic resource allocation gave the system that boost it needed to convince me that it could be a top seller without much effort.

      I don't see a shift to POWER happening for the end user anytime soon, if ever. Intel and AMD can easily beat the price on any POWER chip. But the POWER chips aren't designed to be in your home PC, they're designed to power your server. And that's what they're good at.

    16. Re:Power Rocks by Anonymous Coward · · Score: 0

      Your list of grievances is out of date, especially with respect to the current 64-bit extension of the 4004 instruction set. There are 16 general purpose registers, all of which are equally useful in the fast instruction set, although all the old warts are still there in the compatibility microcode (and many of them are no longer even active in the 64-bit mode). Neither segments nor near and far calls have been used in years, and have been done away with in 64-bit mode. The stack-based floating point instruction set has been replaced by a register-based one. Pipelining and branch prediction are implemented pretty much the same way they are in bona fide RISC processors, and although the decoding stages of the pipeline are more complicated, this complication also gives us a more cache-efficient instruction encoding than pure RISC processors.

    17. Re:Power Rocks by Anonymous Coward · · Score: 0


      Since when are mainframes are a totally different area? The IBM zSeries machines now use the POWER5 chips, and have even better LPAR functionality that is mentioned in the article (more partitions, more processors per system, etc).


      None of the mainframes have used Power/PPC for their PUs. There are ppc processors in the channels, and other subsystems, but the central processors/IFLs/zAAPs/SAPs are not Power/PPC!

      The zseries boxes do have a few more virtualization features. Pseries boxes can have many, many more LPARs than zseries. zSeries boxes are limited to 64 LPARs due to a bit mask (legacy!). pSeries boxes can have up to 640 LPARs. The reason that "thousands" of Linux images can run on a zSeries box is because they are run as guest OSes on top of VM, and guest != LPAR.

      The zSeries boxes max out at 64 cores just like pSeries. In zseries, there are different configuration rules, and the max number of cores for direct use by applications that you can end up with is 54.

      Btw, pSeries boxes can accomodate more memory too.

      I hope someone proofread your book :)

  4. zSeries? by afidel · · Score: 1

    Is this just an extension of the work done for the S/360 nee zSeries LPAR stuff, or was something truely different needed for the POWER line?

    --
    There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
    1. Re:zSeries? by mekkab · · Score: 1

      Different for the POWER 5 chip, because now the Hypervisor runs in firmware but has some hooks into the OS.

      But, yeah. ITs all from the mainframe world.

      I mean, without this you can still run Linux on an LPAR, and you can still give that LPAR fractional CPU allotment (capped or non capped) and frational IO (ethernet, harddisk). Now you can just do it on the fly.

      --
      In the future, I would want to not be isolated from my friends in the Space Station.
  5. Done for years by fiber0pti · · Score: 3, Informative

    This has been available from IBM for years. I've only started with POWER 4 but I think it was available a few versions back as well with limited functionality on Linux. The more advanced features recently came available with the release of the POWER 5 processor. Nearly all of the RAS features are now available for Linux on IBM that have been available for AIX.

    They also support Redhat and SuSE. Good stuff!

    1. Re:Done for years by vdthemyk · · Score: 1

      POWER 4 systems and AIX 5.1 was the first LPAR capable systems. Even 5.1 didn't support DLPAR operations. Anytime you wanted to rearrange your resources, you had to bounce the partitions affected. AIX 5.2 was the first OS to support DLPAR operations. Linux just started to support this with the 2.6 kernel as well.

      --
      VD
    2. Re:Done for years by janic · · Score: 1

      Actually, LPAR on power5 is _quite_ different, if you are used to LPAR on iSeries.

      The pSeries and zSeries guys have had an HMC (HSC, actually) for years. Now that IBM is trying to make it more friendly for the iSeries folks, it actually has a gui. Iv'e talked to quite a few AIX guys that really like the interface, and just as many that want their command line back.

      Of course those guys are very happy to learn that a quick click on the HMC desktop, and you have a shell.

      Anyways... on iSeries power5, you have no os/400 LPAR 0, The enforcement of resource allocation has been taken over by the instance of linux running on the Service Processor. If you have the hardware resources, you can have each Linux LPAR own it's own NIC, disk controller, etc. Otherwise, you can use the VIO server (Virtual IO server, it's a stripped down AIX install) or even use SuSE to provide disk and network.

      But then again, you could use one of your i5/OS partitions to host VIO.

      Same goodies exist now across the entire power line, iSeries, pSeries and OpenPower.

      We've had our i5 for not quite a year now. 7 LPARS, 4 i5/OS, 3 SLES9, 3 out of 4 CPUs enabled and used, and it runs like a champ.

      Personally, what I am waiting for is a POWER5 blade...

      Cheers!
      John.

  6. HOWTO: Slashdot Advertising For FREE by isa-kuruption · · Score: 0, Troll

    New way to advertise on Slashdot, FOR FREE!

    Get a story submitted linking to your corporate website for more information. Then, point to another site, not on your corporate URL, labeled 'Sponsored Content' which once again shows why your product is so great!

    1. Re:HOWTO: Slashdot Advertising For FREE by Anonymous Coward · · Score: 1, Informative

      You know, that's pretty much exactly what I thought when I saw this 'article'. I suppose IBM's LPAR technology on Power is useful, cool tech. But this seems more like a press release than 'news'. This stuff has been around for a few years now (granted, a lot of folks still haven't heard of it, so I suppose in a way it's 'news' to them).

      But, this seems a lot more like an advertisement for IBM than 'news'.

    2. Re:HOWTO: Slashdot Advertising For FREE by turgid · · Score: 1

      IBM has always had some of the best marketing in the world, and it milks slashdot for all its worth.

      --
      Stick Men
    3. Re:HOWTO: Slashdot Advertising For FREE by Hollis · · Score: 1

      What makes you think it was free?

  7. Try using SuSE as a virtual IO server by Anonymous Coward · · Score: 0

    You can share network and scsi adapters
    http://www-128.ibm.com/developerworks/eserver/libr ary/es-susevio/

  8. wait till you figure out SMT! by mekkab · · Score: 3, Informative

    Simultaneous Multi Threading is an even cooler feature of the POWER 5 chip!

    Here's a pdf.

    --
    In the future, I would want to not be isolated from my friends in the Space Station.
    1. Re:wait till you figure out SMT! by Anonymous Coward · · Score: 0

      Also available on Intel's Pentium IV, so you may already have it (and be using it if you have a recent patchlevel of Windows or Linux kernel 2.6 compiled for p-IV).

    2. Re:wait till you figure out SMT! by sapbasisnerd · · Score: 1
      Also available on Intel's Pentium IV

      Not quite the same. While Intel hyperthreading is in broad terms a similar idea the implementation is different.

      As I understand it SMT duplicates all of the stuff before the guts of the processor so there are two complete pipelines etc. whereas Intel has a single pipeline so when the processor switches from one thread to another there is overhead involved in refilling the pipeline from the second thread that does not exist in the POWER implementation.

  9. LPAR on POWER - The Packet Sniffers episode 5 by Anonymous Coward · · Score: 1, Informative

    If you want a better idea of how LPARs are setup on POWER5 hardware check out episode 5 of The Packet Sniffers. They show a 570 system and some brief menu's of the HMC console one uses for LPAR resource management and DLPAR. Not exactly DLPAR on Linux, but the process is the same for Linux as AIX. If you are curious how this works then check it out.
    http://www.packetsniffers.org/

  10. Tip of the iceberg... by vdthemyk · · Score: 3, Insightful

    This document just touches on the capabilities. If you want to see a little bit more detail regarding running linux LPAR's on a POWER 5 system, I suggest heading here.

    This is a good technology, and if there are people wanting to get LPAR capabilities without having to purchase all that extra IBM OS's (AIX, i5/OS), you might look into the OpenPower line. 2 way or 4 way POWER 5 systems that run only linux and can create upto 40 LPAR's on one system. That's bascially like having 40 different Linux servers all running at the same time on 4 total processors.

    I agree this technology has some limitations as of right now, but it may not be a bad idea to look at it. And remember, this is PPC Linux, not your standard Intel Linux. While your boss won't know the difference, you should.

    --
    VD
    1. Re:Tip of the iceberg... by digitalhermit · · Score: 2, Informative

      It's a pretty cool technology. We're running multiple AIX 5.3 partitions on IBM pSeries boxes. Setup of the systems are pretty easy and allocating memory/CPUs is straightforward. The only concern is that the overhead increases substantially with the number of LPARs.

      The only problem I have with pSeries linux is that it somewhat negates the cost advantage of Linux on Intel. Well, make that obliterates the cost advantage. IBMs AIX is free with the hardware. Linux is licensed per instance. So it's cheaper to run AIX *on pSeries* when you have many instances.

    2. Re:Tip of the iceberg... by vdthemyk · · Score: 1

      While AIX isn't "free" (you're charged per active processor), almost all pSeries (soon to be IBM system p5) have AIX included with the system purchase. This does increase the value of using AIX in micropartitions instead of linux.

      --
      VD
    3. Re:Tip of the iceberg... by UncleJosh · · Score: 1

      Or look at some of the articles in the latest issue of the IBM Journal of Research and Development for some more techincal detail.

  11. You must be right.. by scsirob · · Score: 1, Funny

    Apple has just decided to switch to Intel. They are known to pick the underdog technologies in the industry. Power 5 is about to become mainstream, and Intel will become underdog, that's obvious.

    --
    To Terminate, or not to Terminate, that's the question - SCSIROB
  12. Article vs. Advertisement by Anonymous Coward · · Score: 1

    Why is the second link an ad on moving from Solaris to Linux? Any bets that the submitter is really an IBM flunky? Are we going to start putting links to doubleclick pages too?

  13. Xen by SumDog · · Score: 2, Interesting

    This type of technology has been available from IBM for years. I remember those old AS/400 machines during my undergrad that had removable boards that you could hot swap which contained extra processors. One of my professors told me about when he took operating systems, he made his OS on an IBM machine and was able to use one of the six processors available in his own little virtual space without interfering with anyone else's simulations.

    Correct me if I'm wrong, but this seems like IBM has placed into hardware what systems like Xen currently does in software, allocating virtual space for different operating systems to share resources and execute simultaneously.

    1. Re:Xen by rhaig · · Score: 1

      does Xen allow you to change a host OS from having 0.1 of a virtual processor to 2 virtual processors on the fly? I don't know Xen. That's what the dynamic LPAR's do. you can change these things on the fly.

      --
      "We are not tolerant people. We prefer drastically effective solutions"
    2. Re:Xen by Anonymous Coward · · Score: 0

      Yes it does. it partitions on the fly even pci devices, memory, hard disks, ...

    3. Re:Xen by brunson · · Score: 1
      does Xen allow you to change a host OS from having 0.1 of a virtual processor to 2 virtual processors on the fly?


      Not without CPU level hardware support. I don't know if this tiny mention of support means they will have the features required for that level of control, but it's something that piqued my interest.
      --
      09F911029D74E35BD84156C5635688C0
      Jesus loves you, I think you suck
    4. Re:Xen by superpulpsicle · · Score: 1

      Xen is closer to VMware if anything.

      LPAR on AIX and Linux is real virtualization. Similar to Solaris 10's container or the old Sunfire hardware Dynamic Domain Reconfiguration. Split at the hardware level.

    5. Re:Xen by sn00ker · · Score: 1
      Correct me if I'm wrong, but this seems like IBM has placed into hardware what systems like Xen currently does in software, allocating virtual space for different operating systems to share resources and execute simultaneously.
      You're wrong :P
      The reason you're wrong is that the LPAR concept has been in big iron for quite a few years, long pre-dating Xen and even pre-dating VMWare. Saying that IBM have put the Xen concept into hardware makes it sound like IBM are the ones who copied an idea. Xen is a copy of VMWare, with some LPAR-type functionality, and VMWare is a poor copy of the LPAR concept.
      --
      "God, root, what is difference?" - Pitr, userfriendly
  14. Still no support for Dynamic Memory by The+Mad+Duke · · Score: 3, Informative

    Nice write up from IBM, but it's important to remember thaat the Linux kernel only supports dynamic changes in CPU and PCI devices, you can't move memory around. AIX allows dynamic memory; the Linux kernel will need some fundamental chages to enable this. Power5 is indeed the coolest technology around today, but Dynamic LPAR started on the Power4 back in 2001, so this is kinda old news.

    --
    -The Mad Duke
    1. Re:Still no support for Dynamic Memory by Anonymous Coward · · Score: 0

      DLPAR on Linux isn't old news, however. Also, dynamic memory support is on its way.

    2. Re:Still no support for Dynamic Memory by Chanc_Gorkon · · Score: 1

      This is more then enough for me to stick with AIX. Granted, lots of times the LPAR's Never change, but if you had spare memory and you want to allocate it without taking it down, you can't do it with Linux...yet. AIX is nice like that....

      --

      Gorkman

    3. Re:Still no support for Dynamic Memory by kpharmer · · Score: 1

      I've got a p570 sitting in a data center waiting for its install right now. I'm going to be using AIX 5.3 and DB2 8.2.

      Linux would be fine, but there's no price advantage, and aix is more mature. But dynamic memory isn't the real issue there - most of the memory will be consumed by the database - so it's db2's ability to dynamically change memory footprint that is the more critical.

  15. nanokernel by Doc+Ruby · · Score: 1

    I'd love to see a version of "Linux" that executed functions solely to allocate system resources. Authenticated access by processes to each other and to the hardware that some of them represent (drivers). All other threads/processes would be userland apps. This LPAR system would offer enough flexibility under the hood of the actual OS that the rest of the system could be highly efficient, while also simple, secure and distributable.

    --

    --
    make install -not war

    1. Re:nanokernel by vdthemyk · · Score: 1

      That's exactly what the Virtual I/O Server is. You can get it with any p5 system that has APV (Advanced Power Virtualization). However, it's a skinnied down version of AIX instead of Linux.

      The VIOS acts as a broker of disk and communication resources. You can have one ethernet adapter assigned to a VIOS and create upto 20 virtual LAN's and basically an unlimited number of SEA's (Shared Ethernet Adapters) for client partitions.

      As for disk, you can have one SCSI controller and create vSCSI adapters on the client partitions so the clients can utilize the same SCSI controller as other partitions. You can break it down to the device level (each pysical disk) or to the lv level (a section of a volume group).

      For those of you who think this is just an IBM thing, I hear Intel is planning to support virtualization shortly as well for Windows systems. AMD is also on the bandwagon. Not to mention that HP has vPAR's and nPAR's and Sun is working at it a different way with containers.

      Bottom line is virtualization is getting huge fast. It's been proven to decrease cost and increase system utilization and availability.

      --
      VD
  16. Seems cool by Anonymous Coward · · Score: 0

    The ibm speaker at the ohio linux confrence last weekend talked about this. It seemed damn cool. Too bad we are all sun here atm.

  17. How long before.. by Hesperus · · Score: 1

    .. SCOG claims this as having been stolen from the code they think they own?

    --
    ____________________________________

    -- I beleve you'll like this -->
  18. Linux Needs This by olddotter · · Score: 1

    Since logical partitions is the only cool technology I have heard of that is new in Solaris 10, Linux needs this LPAR support (everywhere) to keep making in roads in Fortune 500 comapanies' datacenters.

    --
    Think Deeply. ...
    1. Re:Linux Needs This by Anonymous Coward · · Score: 0

      "Since logical partitions is the only cool technology I have heard of that is new in Solaris 10, Linux needs this LPAR support (everywhere) to keep making in roads in Fortune 500 comapanies' datacenters."

      Sun needs processors with competitive performance in order to keep making in roads in the Fortune 500 comapanies' datacenters. I.e. (from yahoo biz):

      Sun has also proved additional compelling performance comparisons against competitors:

              -- On SPECjbb2000 the 24-way Sun Fire E6900 outperforms 32-way Itanium-
                        based HP servers, a 32-way Fujitsu PRIMEPOWER 1500, and a 16-way IBM
                        p5-570.
              -- For scientific computing on the LINPACK "N" benchmark, the Sun Fire
                        E6900 also outperformed a 1.9GHz IBM p5-570 by 22%.

      The term "way" is by no means a consistent unit across these platforms. Let me clarify:

      24-way Sun Fire E6900 = 48 cores
      32-way Itanium-based HP = 32 cores
      32-way Fujitsu PRIMEPOWER 1500 = 32 cores
      16-way IBM p5-570 = 16 cores

      Sun has said that their main competition is Power, but they can't seem to beat Power in any benchmarks. The first Power5 boxes showed up over a year ago! You'd think they'd be embarrassed, in the case of SPECjbb2000, to say "look, our brand new machine with 3 times as many cores, beats your 1+ year old machine.". WTF, it better!

      Maybe I'm not familiar enough with Sun hardware. Maybe the Sun Fire E6900 is only a 4U machine. In that case, Sun should have put their emphasis on floor space saved, not on raw performance!

      On the other benchmarks, check out the latencies.

  19. Solaris Zones. by Anonymous Coward · · Score: 0

    This is different then Solaris (SPARC/X86) zones how?

    1. Re:Solaris Zones. by chez69 · · Score: 1

      it is at the hardware level instead of the os level

      --
      PHP is the solution of choice for relaying mysql errors to web users.
    2. Re:Solaris Zones. by Anonymous Coward · · Score: 0

      Of course, you can do this stuff on Sun kit too.. On a 15/25K or a 1280-6900.
      I'm not sure how they compare (for Sun hardware, your hardware is limitted per uniboard or pci board, for example you can't assign 2 cpu's to one domain, and two to another; from the same SB)

      But in general it seems to be similar.

      Dynamic Reconfiuration (what it's called in the Sun world..) has been around for several years now, works very well and is very stable. It's mostly in the firmware of the the boards though (check cfgadm in OpenSolaris)

    3. Re:Solaris Zones. by chez69 · · Score: 1

      Cool, I thought sun kit had to do it in software because because the x86 version of solaris does zones.

      good job sun.

      --
      PHP is the solution of choice for relaying mysql errors to web users.
    4. Re:Solaris Zones. by Anonymous Coward · · Score: 0

      Yes, Zones are in Solaris 10. What Zones do are pretend to be a unique instance of Solaris, with its own filesystems, ip addresses, etc under one global zone. However, all the Zones share the same physical hardware.

      What DR achieves, is you can dynamically add/remove memory/system boards, io boards, to Solaris (on the hardware platforms that support it, see my original post). If a DIMM or CPU needs replacing, you can simply replace the board, swap whatever needs to be swapped; and then pop the board back into the domain, whilst Solaris is still running. Or you can shuffle boards around depending on load, maybe from from 9-5 database A needs 24 CPU's but aftre 6 becomes idle, but database gets more load, so you can take 2 boards from Dom A and pop them into domain B, adding 8 CPU's.. You get the idea.

    5. Re:Solaris Zones. by Anonymous Coward · · Score: 0

      I have a friend who works on big sun hardware who told me all about it. /me drools

      Now I'm waiting for this to hit x86 so I can run a linux and windows xp simultaneously. It'd be nice if I could dynamically reallocate resources to each system while they're running and them be aware, so I could give one more cpus or memory if I'm doing something more intensive atm on one system while the other is more idle... oh well... someday maybe.... or maybe by then the pc will be my cellphone... keep dreaming.

  20. LIcensing? by Necron69 · · Score: 2, Interesting

    I found this interesting. Try this on the bits for the RHEL4 packages:

    for i in *.rpm; do rpm -qilp $i |grep -i license; done

    Size: 2547252 License: IBM Corp.
    (lots of output deleted)

    11 "IBM Corp." licenses.
    1 IBM Common Public License (CPL)
    1 GPL

    Open Source it ain't. I'd rather use Xen.

    - Necron69

    1. Re:LIcensing? by Anonymous Coward · · Score: 0

      LPAR, DLPAR, and Micro Partitions (Power 5 only) are implemented in the IBM hardware (The hypervisor to be exact). They are not packages that ship with your distribution.

      Some of the features of this may be enabled (or utilized) through software (read: your distribution), but you are talking about open source software as opposed to a system architecture.

  21. Re:Done for years - not really by Anonymous Coward · · Score: 0

    LPARs have been around since the Mainframe days... on s/390 technology. LPARs on non-mainframes are relatively new. This concept was first introduced with the POWER4 technology (think p690s etc), in which you could allocate resources to the different partitions in your system. DLPARs is new to the POWER5, using newer AIX 5.2/Linux 2.6.... micropartitionins/virtual io is something really nifty from IBMs part. Run 10 servers (independent of each other) on one single processor... and you can still dynamically change them... drool... :D

  22. Good stuff... by Chanc_Gorkon · · Score: 1

    We currently have a 570 and it's awesome. We can allocate resources on the fly to any of our AIX partitions and we can also run Linux on a LPAR and AIX on a LPAR and even OS/400 (or what ever they call it now as AS/400's are now basically Power Machines). DLPARing let's you allocate memory, processors and disk from one partition to another with out need to take it down. IBM makes THE BEST hardware around...BAR NONE when it comes to reliability and availablity. It is GREAT stuff.

    --

    Gorkman

  23. Does this work with PowerLinux? by ubiquitin · · Score: 1

    Not sure about Linux on Power, but how about PowerLinux? That's what I use.

    --
    http://tinyurl.com/4ny52
    1. Re:Does this work with PowerLinux? by vdthemyk · · Score: 1

      If the linux kernel used is v2.6, you'll probably be ok. However, IBM won't help you out if it doesn't work. From what I understand, it has to be SLES 9 or RHEL 4.

      --
      VD
  24. partitioning vs cheap hardware by markhahn · · Score: 1

    it's possible to take server consolidation too far. suppose it costs you $2e6 to buy an IBM mainframe that can support 200 LPARs (I mean real, active ones, not idle ones.) when is this better than putting each on a $2k server of its own? sure, sometimes it is: the LPAR can react more dynamically, and some aspects of TCO would be lower. but we have to be honest when making this comparison - let's assume the separate servers are auto-provisioned, for instance, and have IPMI and some sort of intelligent storage network. the operating costs of such an alternative is fairly low, perhaps lower than IBM's relatively exotic hardware.

    don't get me wrong: resource pooling is a great thing (it's what I do, come to think of it), but the kind of partitioning that IBM is pushing is, IMO, more of an effort to sell high-end hardware. and doing a really honest/complete TCO analysis is quite complicated. virtualization is not a value in and of itself.

    1. Re:partitioning vs cheap hardware by vdthemyk · · Score: 1

      I agree to a point. Some situations it makes more sense to fill up a rack with 1U 2way intel servers and run windows or linux as an OS. However, if some of the servers are running at 10-20% processor utilization and others are getting maxed out on processor utilization, how long does it take to redistribute your processing resouces? With virtualization, provided you have the processors assigned to a pool, the system automatically will move processing power from one LPAR that is not maxed on processor resources to an LPAR with high utilization. This happens every clock cycle. And if you run into a situation where this is happening often, you can go ahead and move over as small as a .01 of a processor from one LPAR to another. This grainularity can greatly increase your processor utlization.

      Lets extend this to I/O resources. If you have a system that has a high I/O through put on the LAN and another system that has very low throughput, there isn't a way to balance that load is there? With a VIOS you can have as few as one ethernet adapter provide LAN connections for 100's of client partitions. Yes, this can put an increadible ammount of load on that one adapter, but if you have say 10 Gb LAN adapters assigned to the VIOS, you can aggregate them into one pipe, and carve that logical enet adapter into 100's of Shared Ethernet adapters across 20 VLAN's for the client partitions. This can get you to the point where your I/O bandwidth is spread across multiple LPAR's all while maintaining secure networks for each partition.

      --
      VD
    2. Re:partitioning vs cheap hardware by kpharmer · · Score: 1

      I think an even more compelling situation is one in which you max out your 2CPU or 4CPU box: and have no where to go for extra cycles/bandwidth/etc.

      In a lpar situation you can still easily provide the extra resources - with a more traditional server configuration you'll have to rehost. Which will completely waste the cost of your prior server, plus you've got to pay for a larger one as well - which may only be maxed out once a month. Then there's the labor, outage, and risk involved in the migration - which can easily exceed the cost of the hardware itself.

  25. LPARS by bigredradio · · Score: 1

    This has been around for while. Did anonymous coward just now find out about it? Or is anonymous coward IBM marketing just trying to get the word out. Anyway, I have been using it for about two years. It's pretty impressive technology. The recent addition to sub-processor partitioning is really cool. However, the one item that seems out of place is the fact that you need a separate system (a Hardware Maintenance Console HMC) to manage it. This is a separate linux x86 box ($4,000 beige box) that you have to dedicate to manage the LPARS. If the LPAR management could be included in standalone system, it would make more sense. Other than that, it's impressive technology.

    --
    Flexible bare-metal recovery for Linux/UNIX
    1. Re:LPARS by vdthemyk · · Score: 1

      I agree. The HMC is definitely a bad part of the LPAR management. It will be interesting if IBM addresses this with the new POWER5+ launch slated for the next few weeks.

      But the HMC does provide a very good way of supporting multiple systems from one console. I'm not sure if you're working with p4 or p5 architecture, but on the p5 they've moved it from a serial network to an IP network for managing the servers. Much better to have one CAT 5 cable out of that HMC than to have a few RAN's hanging off the HMC attached to servers directly.

      The thing I like the most about the HMC is the firmware updates for the managed systems. Makes it very easy to do from one console.

      --
      VD
    2. Re:LPARS by cgh4be · · Score: 1

      Check the IBM pSeries website on October 4th. I think you'll be pleasantly surprised :)

    3. Re:LPARS by ThatDamnMurphyGuy · · Score: 1

      It just so happens that we had a guy talking about the new Power5/LPAR/VM stuff at the Ohio LinuxFest Saturday. I won't claim to know very much about the topic, but the presentation was very clear about that fact that the new VM/Virtualization stuff in Power5/Linux isn't you're fathers LPAR. While you can have up to 40 LPARS, you can have many many more VM servers on top of that, or even on top of the hardware without LPARS at all.

      I'm trying to find the powerpoint of from the speaker in question (Scott Courtney, Sine Nomine Associates). It had some interesting number on the max number of Linux VMs they could run befure it ran out of resources.

    4. Re:LPARS by as400tek · · Score: 1

      I was apart of the first bunch or regulars to get to look at LPARs on the iSeries back in 2000. Thay are great. The AiX group gave us HMC and as a side note my brand new 570 with the HMC showed up this AM and man I am giddy!

      --
      David Vasta iSeries(AS/400) Admin & Junkie
    5. Re:LPARS by Anonymous Coward · · Score: 0

      "Did anonymous coward just now find out about it? Or is anonymous coward IBM marketing just trying to get the word out. "

      I would imagine the way it works is some shill tries to get an editor to post the article for free, but the editor will offer to launder^H^H^H^H hide the shill's identity for a fee.

    6. Re:LPARS by vdthemyk · · Score: 1

      bigredradio,

      I thought you would be very interested in this. IBM just announced the IVM. It allows you to partition a single server without the need of an HMC. That functionallity has been moved internal to the system. Check out the latest announcement.

      --
      VD
  26. Addendum (same AC) by Anonymous Coward · · Score: 0

    It turns out (in pIV) to be no more efficient than just running two tasks time-shared on an Athlon.

  27. Re:hey guys... by as400tek · · Score: 1

    I wonder if you could have put a liittle more thought into that comment. I woudl agree with you on some points, but I am not sure which parts you don't like.

    --
    David Vasta iSeries(AS/400) Admin & Junkie
  28. Old NEWS! by JhohannaVH · · Score: 1

    Heheeh... I've been doing this in production for a year now! It's just NOW making it to Slashdot? That's frickin' funny.

    In fact, I'm running Linux & AIX on the same cluster, separate LPARs. All over the place. Hrmph. Jho

    --
    Sorry man... the Internet pooped on me.
  29. One other intresting note.... by Chanc_Gorkon · · Score: 1

    The HMC on a Power 5 machine uses a modified Linux distro and also uses Fluxbox for the Window Manager. Good stuff.

    --

    Gorkman

  30. IBM's hidden Charges by Anonymous Coward · · Score: 0

    I don't mean to rain on your parade... but IBM hides a lot...

    IBM will typically try to sell you a $4k machine dual proc that can run say 20 LPARS. So you buy one quickly since you all of the sudden can consolidate!

    On top of that, if you run the maximum LPARS you lose about 30-50% of your performance on LPAR switching overhead.

    To top it all off, each LPAR is licensed separately. So it's like $1500 per LPAR turning your $4K box into a $34K box that runs as fast as a $2k box. All of the sudden, this LPAR joke doesn't sound so good any more does it?

    Sun realized this and went for Zones (glorified Chroot jails) and it works fairly well. The downside is that all of the sudden patching software in the global zone is a bit tricky, so it's not perfect. However, you won't see a crazy license bill coming from Sun...

    Anyways, just beware... IBM has really ticked me off with their marketing... and I'm one of their resellers! Come ask me about COD one day... they tell you that it's free! You don't pay for the HW till you use it! Except what they don't tell you is that you _DO_ have to license the priviledge to have COD... So in the end, you pay more... and BTW the fastest Power procs only come COD... so first you gotta pay COD, then you gotta pay to activate it... Ugh!

  31. It's all about management by Anonymous Coward · · Score: 1, Interesting

    AIX and power hardware has always been about management and stability not nifty features.

    What about the concurrent firmware upgrade feature that has just been rolled out as promised in the latest power5/HMC code

    - anyone else ever tried firmware updating 7 boxes some running multiple partitions at once WITHOUT disruption to the patitions in question?

    I just have - nice little feature!

  32. You forgot the $4k HMC that you need, by redwoodtree · · Score: 1

    Right?

    1. Re:You forgot the $4k HMC that you need, by Anonymous Coward · · Score: 0

      Not anymore. See the Integrated Virtualization Manager redbook:

      http://www.redbooks.ibm.com/redpieces/abstracts/RE DP4061.html?Open

  33. What is up with linking to an advertisment? by Anonymous Coward · · Score: 0

    That is pretty lame, linking to an add for IBM migration services.