1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

News Russian company promises x86-on-ARM emulation

Discussion in 'Article Discussion' started by Gareth Halfacree, 4 Oct 2012.

  1. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    12,401
    Likes Received:
    1,814
  2. Guinevere

    Guinevere Mega Mom

    Joined:
    8 May 2010
    Posts:
    2,477
    Likes Received:
    170
    A 20% hit to efficiency isn't bad, but if they're taking a 60% dive at the moment then they have a LOT to do to get that far.

    So what's the final aim for this? To run an x86 OS as a VM on ARM hardware? Is there really much demand for it? Maybe in a few years. I can't see enterprise making the shift to ARM servers while still wanting to run their legacy OSs on them, not when there are so many options with hosting x86 software on x86 hardware.

    A few years time (if it happens) when there's a greater reason to have ARM servers I can see being able to host legacy OS versions a plus.

    The other thing you could do with this (I guess) is host a full x86 OS on an ARM tablet (A rooted android, win8 RT... even an iOS device). Slow performance and terrible battery life but it could work...

    If this software layer does what they claim.
     
    Last edited: 4 Oct 2012
  3. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    12,401
    Likes Received:
    1,814
    No, it's to run native x86 applications on ARM. They're targeting the datacentre. Let's say, for example, that a few years ago you paid to have SuperFinancialTransactionServerPro written for you. It runs fine on your x86 servers, but now you want to move to ARM. Uh-oh: SuperFinancialTransactionServerPro doesn't run on ARM. Solution: buy the ARM servers anyway, run SuperFinancialTransactionServerPro under emulation, and eventually port the software across to ARM native.

    Look at Rosetta on Mac: when Apple moved from PowerPC to x86, it was able to write a new OS no problem. Trouble is, none of the client software everyone had would run. Companies like Adobe promised to port the software to x86, but it wasn't ready - so Apple hired a company to write an emulation layer that allowed PowerPC applications to run on x86 hardware under an x86 OS. When there had been Intel-native versions of all the apps out for a few generations, and the old PowerPC Macs had fallen well out of extended support contracts, Rosetta - which was only ever a stop-gap measure - was killed off. This company is doing exactly that, but swapping 'x86' for 'ARM' and 'PowerPC' for 'x86.'
     
  4. Guinevere

    Guinevere Mega Mom

    Joined:
    8 May 2010
    Posts:
    2,477
    Likes Received:
    170
    Yes I get that. But what OS does "SuperFinancialTransactionServerPro" run on? The EE article refers to "server software" which could be something like "BlackBerry Enterprise Server" that runs above windows or "VMware ESX" that runs bare metal.

    If it's bare metal hardware, then yes maybe it'll run onto of his x86 hardware virtualisation layer, but very little is written to work like that and isn't that just hosting an OS in a VM anyway? So much server needs need a full windows OS and about half a dozen windows API flavours or a full linux behind the scenes, which is why I was wondering if the aim was to go with full machine virtualisation.

    It's possible with russian devs are hoping their code is adopted to be adopted by an OS vendor and used like rosetta to run pre-compiled code but it isn't clear from this BT article or the source if this is the case.

    Which is why I asked.
     
  5. dyzophoria

    dyzophoria Member

    Joined:
    3 May 2004
    Posts:
    391
    Likes Received:
    1
    do we really need to run x86 stuffs on ARM? I kinda like running native software in ARM :D only..
     
  6. azazel1024

    azazel1024 New Member

    Joined:
    3 Jun 2010
    Posts:
    487
    Likes Received:
    10
    What kind of server response times are you likely to be looking at though in scenarios where you have high density, low performance per core servers? I certainly think that has its place, but I'd imagine that isn't going to fill the needs of a lot of people/companies and a 20% performance penalty on top of that...

    Performance per watt is pretty useful, but it can also be a pretty poor goal in mind when other things like server response time, total FLOPS, memory bandwidth, memory density, overall build power budget, etc need to be considered as well.

    Great for low performance micro servers and clusters. Not so good for medium/high end applications. Also with a 20% performance hit goal...that might not stack up so well against upcoming Atom designs which aren't going to have that performance hit, and radically better processing capabilities and efficiency compared to current atom designs from everything coming out of Intel.
     
  7. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    12,401
    Likes Received:
    1,814
    Whatever OS it ran on before, or a compatible OS. Remember, the overwhelming majority of server applications - especially in the many-core highly-parallel market - run Linux or another Unix-alike. In the case of Debian, for example, you can get the latest version in ARM, MIPS, PowerPC, x86, x86-64, and a dozen other flavours. Buy an ARM server, install the ARM build of Debian, install the Russian x86-to-ARM compatibility layer, install your x86 build of SuperFinancialTransactionServerPro. Job's a good 'un.

    If you were looking at Windows apps - which are a minority in the datacentre world - then you could just use wine to provide the Windows compatibility needed to have it work under Linux. Bosh.

    EDIT: That's unlikely to be the case, though. Remember that the target audience for this product is people already thinking about moving to an ARM platform. With no current generation Windows server product on ARM, it's highly unlikely that anyone relying on a Windows Server environment is even considering moving to ARM.

    The OS isn't the problem; the application is the problem.
     
    Last edited: 4 Oct 2012
  8. dicobalt

    dicobalt New Member

    Joined:
    21 Mar 2009
    Posts:
    169
    Likes Received:
    2
    A 20% penalty while emulating x86 on ARM is totally believable and I have a bridge for sale.

    It'll never ever happen unless they modify the ARM instruction set, but then it wouldn't be ARM anymore.
     
  9. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    12,401
    Likes Received:
    1,814
    I have no trouble believing it, if they're talking about ARMv8 - the hardware virtualisation extensions are going to help immeasurably.
     
  10. Anakha

    Anakha Member

    Joined:
    6 Sep 2002
    Posts:
    587
    Likes Received:
    7
    Wow. They discovered Bochs. Or perhaps qemu
     
    Last edited: 4 Oct 2012
  11. faugusztin

    faugusztin I *am* the guy with two left hands

    Joined:
    11 Aug 2008
    Posts:
    6,859
    Likes Received:
    247
    I think this whole project is a result of the Elbrus 2000 "fiasco", they took the only usable part of that project :
     
  12. Gareth Halfacree

    Gareth Halfacree WIIGII! Staff Administrator Super Moderator Moderator

    Joined:
    4 Dec 2007
    Posts:
    12,401
    Likes Received:
    1,814
    Not exactly: neither Bochs nor qemu - correct me if I'm wrong - will allow you to execute x86 applications within a non-x86 OS. Instead, they allow you to run an x86 OS on top of your non-x86 OS, then run the application on that - something that will have far higher overheads than the claimed 1MB of memory needed for Elbrus' system.
     
  13. AmEv

    AmEv Meow meow. See yall in 2-ish years!

    Joined:
    6 Apr 2011
    Posts:
    1,173
    Likes Received:
    43
  14. mdshann

    mdshann New Member

    Joined:
    18 Nov 2010
    Posts:
    56
    Likes Received:
    0
  15. Bakes

    Bakes New Member

    Joined:
    4 Jun 2010
    Posts:
    886
    Likes Received:
    17
    To people who seem skeptical, I believe mainframe emulators were used for this sort of thing for quite a while (many are probably still in use) to allow old COBOL software to be run on newer, non-broken/bespoke hardware.
     
  16. trevj

    trevj New Member

    Joined:
    5 Oct 2012
    Posts:
    1
    Likes Received:
    0
    Minor correction: The RiscPC was designed around accepting a separate physical x86 processor card. Software is also needed to integrate it with the host ARM processor and OS.

    The fully software solution was a forerunner to this on earlier machines.
     

Share This Page