Multiple OSs Concurrently

Josh H. writes "VMware will said they will debut a product at Demo99 that lets the user run multiple OS's here. If this thing really works like they says it does, this could means no more reboots to change OS's for those of us running dual-boots."

Comments

a great idea

[Anonymous Coward]

For $300, I'll reboot

How do you switch?

[Anonymous Coward]

Special Keystroke to switch between OSes?

Do they run concurrently, or does one OS suspend when you use another?

And why is there a separate Windows and Linux version if they run multiple OSes?

wow -- notice the headline though?

[Anonymous Coward]

It's funny how the headline tells about using the extra space around the screen, not about the multiple OSes. Sick, eh? :D

But it sounds really cool. If it works well, then I'd be happy to pay $300. Although I don't understand what good a Linux-only version would be. Run multiple bootups of Linux on the same system? Why? It's not like it's gonna crash anyway.

How about a FreeBSD and NT version? That'd let me run my normal stuff in Linux or FreeBSD, but quickly switchover to play a game in 95 (or NT, if it supports NT).

Oohh, another possibility. =) Load this up on an office servers.. that quad xeon sitting over there.. play Quake 3 (taking advantage of multiple CPUs) in one OSes, then, doh! the boss is coming! quickly toggle over to the real OS (FreeBSD or Linux). Just working on the server boss! ;)

"This is possible because each VM has ..."

[Anonymous Coward]

Hi All,
It sounds interesting, but I'm not sure that it's possible to do what they're implying. What they're actually doing *seems* to be different from what they're implying.
That is, it's my understanding that x86 chips can only have Virtual Machines which emulate an 8086 (or 80286?). They can't emulate a 386. Thus such emulation would have to be done in software if you wanted to run a different 386 OS inside an existing 386 OS. That's what makes running e.g. Win95 inside OS/2 nearly impossible - they each want to control the basic chip. If Intel had enabled V86 mode to do 386 emulation, then things would be a *lot* easier for those of us who want to run multiple OSes simultaneously.
VMWare says that VMWare is based on Virtual Platform, "runs directly on the processor wherever possible to achieve maximum performance" - a bit of wiggle room it seems to me.
On another page they say, "VMware Virtual Platform allows virtual machines to work in concert with each other, sharing files and devices. This is possible because each virtual machine has its own unique network address." It sounds like VP is acting rather like a network redirector. It doesn't sound like they've got a fundamental solution to the lack of hardware 386 emulation on Intel chips.
If the product works as they describe, it'll be very interesting to see its performance. Has anyone seen VP running?

Simos

[Anonymous Coward]

I was quite intrigued that this had suddenly stepped out from nowhere. Writing something like this had been on my wishlist for some time. I did a bit of digging and came up with this:


http://simos.stanford.edu/

The same people who developed SimOS, formed a company to commercialise vmware.

I'm quite sure it is a real product. The chief scientist is a heavyweight OS researcher who developed SimOS. Also linuxcare have given a testimonial saying how useful it is in a multiple Linux environment.

I say good luck to them. It would be nice if there were some differential pricing mechanisms. I'd use it, but only to play with, and it's not worth $300 for that. I'd imagine that it will give Interix some sleepless nights too (overpriced crap for NT).

Paul

One company controls the hardware?

[Anonymous Coward]

What about when you get new hardware VMware does not support? Hardware not supported: Ooopps, I need to talk to VWware. What do you mean It will be 6mths???

Personally I would buy two machines. You than have no contention or software incompatabilities. The way M$ works this would open a battle ground like CORE WARS. Intel based machines/cards/periphirals are not really designed to handle shared resources.

my two cents

[Anonymous Coward]

Microkernel based operating systems have been around for a while, but they simply don't have the necessary efficiency to be competitive with traditional monolithic designs.

Which might not always be the case; see, e.g., the Flux project.

Catch the bottom?

[Anonymous Coward]

Notice the blurb about the other program demo that makes a toolbar in the extra video memory outside what Windows uses? Just what we need, another patent. Let's make sure XFree86 uses all the memory space it can, shut them out early. :)

I've seen it running

[Anonymous Coward]

I've seen the beast this weekend - I have some friend working there and he made me a demo on his laptop computer.

I've seen their virtual machine running in an X11 window and booting NT, then running word. NT really works like if it was running in a "real" computer.

They virtualize all the hardware : the system you boot in the VM will not be able to poke in your real hardware's registers, but when it pokes into
the virtual hardware's registers it get intercepted by the VMware system, which figures out what the request is (send a packet on the net, read a disk sector, etc) and translates the request to the real OS. The emulated hardware needs not be the same one as what you physically own.

The system was slower than a "real" NT box, but it was still useable. I dont know how fast NT natively runs on this particular laptop machine, so I cannot give speed figures, but it was at least 50% of the real speed, probably even more. So this is definitely not some bochs-style emulation here. The NT boot sequence was particularly slow though, probably because NT is probing for lots of hardware and it slows down the emulation ????

I do not know how they achieve this. My guess is that it is some mix of code verification, to check that there are no difficult-to-emulate things like when the OS modifies its page tables and such, and native execution, but I do not know anything more.

Anyway, its a very nice product and I can't wait to play with the beta. $300 is a bit expensive, but I may be able to justify it at work. well at least I plan to :)

hey

[Anonymous Coward]

Okay technically.. something like this IS very posible... but part of this is threw emulation. You cant have FULL prossesor access to ring 0 when another os is using it. Two examples.. dosemu and bochs.. Dosemu is VERY VERY FAST... it mostly translates, not emulates.. but you cant run full protected mode software under it (win95, or most anything else) Because that software requires control over execution ring 0 and the existing os allready has control... Bochs on the other hand TOTALLY emulates the prossesor (well there working on making it relay most of the recieved instructions to the real prossesor like dosemu.. but untill then its going to be SLOW, but atleast using TOTAL emulation you can run bochs on just about ANY prossesor... hehe)... now some sort of cross between the two, that is well optimized could then run an os at nearly full speed.. and even take advantage of any of the controling os's features (ex: network, filesystem.. etc etc.. running your unix daemons WHILE playing tomb raider :)

its all posible..

but ONE thing that gets to me..
ANYONE remember Dosix? I think thats what it was called.. it was supposed to do the EXACT same thing as what we see in this article... they had a beta testing thing where you can sign up but for some reason they NEVER accepted anyone.. progress seemed REALLY slow... fow such an awsome tool it didnt have any real signifigant url (like www.dosix.com hehe) you would think that they would either setup some sort of domain or someone would donate it to them.. such a thing would only change things as we know it.. thats all :)... anyhow.. anyone also notice how it DISAPEARED ALL OF A SUDEN.. web pages removed... all traces of existance GONE..

(shrug) I dont know... you tell me.. all I know is its 8:11 am and Ive only had about 2 hours sleep.. hehe sorry for my ramblings :)

Been There - Done That

[Anonymous Coward]

IBM has been doing this for many years with their VM/CMS operating system. Each VM 'account' is its own virtual machine. Any software running in a VM account thinks that it's running on the real hardware, not in a virtual machine. It's pretty cool because you can test installation of new versions of the VM operating system within a virtual machine. When everything's working, switch over you can then reboot the machine using the new version of VM. Many sites also run several instances of MVS under different VM accounts. It's a good way to separate test from production. That way, if MVS crashes, it only takes down the VM account and not the whole machine.

Sounds a bit like Mach

[Anonymous Coward]

This sounds similar to a rumor I've heard quite some time ago about a software developer at Apple having MkLinux and Rhapsody (AKA MacOSX) running concurrently on the same Mach microkernel. Of course that was PPC hardware and both OSes were written to run on some kind of HAL.

This seems feasable for such a product to make it to market but I wonder about the overhead. I have heard of people benchmarking MkLinux vs. LinuxPPC (microkernel vs. monolithic) and have shown 5% to 20% speed differences. OTOH, Sheepshaver (www.sheepshaver.com) claims no loss of performance when running MacOS inside a sheepshaver window on BeOS.

In short, very cool, very workable, but at what performance loss?

Kurt Sellner
ksellner@earthlibk.net

Multiple OSs Concurrently

[Anonymous Coward]

Isn't this what (most) micro-kernels have been
doing for a long time ??? For example, with MACH
micro-kernel, you can perfectly have several
linux kernels active at the same time, even on
Intel hardware ... I don't know about NT, but
why not, if somebody did some serious reverse
engineering ?

NB. with a micro-kernel (e.g. Mach), AFAIK, the
micro-kernel manages the physical (real-world)
hardware including interrupts, etc... much like
IBM's VM did already more than 30 years ago !!!

How do you switch?

[Anonymous Coward]

Ouch, if it runs on top of one of the OS's (that's wierd) it sounds like you want the Linux version and not the NT version. But will it do BeOS, or multiple processor systems?

GOD --- LEARN TO READ!!

[Anonymous Coward]

Please, learn to Read!!.. this is a software ___FOR___
Windows NT and Linux thats run x86 REAL MODE
software (like win3.1 and dos) ..... is like DOSEMU

ahahahhaa.. read it again! :)

Is this new?

[Anonymous Coward]

Sure sounds like VM/360 to me. For all that goes they did it in the early 90's with the Mach microkernel running OS/2 and AIX (the x86 version, remember that). Now that was a SWEET setup! To bad it was scrapped.

Order of comments

[Anonymous Coward]

hehehe. anyone notice how the order of the comments are all mangled on this page.
So Am I first now? hehe

Yes but...

[Anonymous Coward]

But it's not sexy to say that IBM has been doing this for decades. After all, VM is old stale technology that doesn't work as well as today's newer operating systems. Just like clustering is sexy but parallel sysplex is boring.

I wonder how long it will be until PCs can set up LPARS.

As the song goes "Everything old is new again..."

uh vm86() anybody?

[Anonymous Coward]

Linux already has sycall vm86(). It's what dosemu uses, and dosemu supports DPMI as well. You could easily hack up a boot loader, an image file of whatever 16 bit OS you want to run, and vm86() it.

And for NT? What crack are they smoking? It already runs dos and win 3.1 programs.

Nothing new here, and definately not worth $300.

Ideas

[Anonymous Coward]

This has some cool uses. Each OS could be behind a firewall OS like this.
(Linux W/ firewall)--------Internet | | NT-+-OS/2 98-+-DOS | Solaris Handy for us admins!

Virtualize the hardware.

[Anonymous Coward]

The trickiest part of creating a VM on a PC is probably the peripheral hardware. Basically, for each device in the PC, you've got to create a virtual device on the virtual machine. You can just write one virtual device for each class of devices: NIC, video, sound, drives, drive controllers, modem, BIOS, PIC, parallel, serial, keyboard, mouse, etc. (For example, they'd probably just make every network card look like an NE2000 and every sound card look like a SoundBlaster 16.) But the virtual device has to look exactly like actual hardware to every possible guest OS. And that's not an easy thing to do at all. (Note that they've licensed an actual commercial BIOS, so that part should work pretty well.)

So the host OS would handle interrupts on the sound card, but if the guest OS is using the sound card at the time, VMWare (via the virtual device) would have to pass on the interrupt to the guest OS.

Because DOSEmu only has to run one guest OS (although of differing versions), it can use a different technique for many devices. It installs a device driver on the guest OS that knows that it is running on a virtual machine, and can call back to resource-handling routines specific to the VM. This is a better solution, but limits the possible guest OS choices. VMware claims to work on too many guest OSes for it to have implemented this feature.

booch (no password -- at work)

You can do this with Mach...

[Anonymous Coward]

I'm pretty sure you can already do this with OSes that run on the Mach microkernel.

For example, I'm pretty sure you can boot multiple copies of MkLinux simultaneously on PowerPC hardware... good for debugging purposes.

Microsoft May Kill it

[Anonymous Coward]

Microsoft does not want anybody to run
the legacy OS like Linux. So MS will probably
buy VMware and its patents for $999M
and then trash it.

This won't hurt a bit.

[Anonymous Coward]

If this system makes it easier to _not_ port software to linux, then over time the platform weakens. WINE, VMWare, etc. do not help the platform in the long run. Only native ports of useful software will do that.

I agree that making it easier not port software is harmful (look at OS/2.. "it runs windows so why develop a native OS/2 port?"). However, in this case I don't think it will hurt because VMWare goes both ways. In this case, it may be easier for a company to develop a Unix version (since Unix is a great dev. environment) with the reasoning that Windows people can still use it if they use VMWare..

I think this is only good news.. Can't wait to get my copy, if it works as planned!

-Tom

$299?

[Anonymous Coward]

For $300, I think I can reboot. I am curious however, about the overhead involved with switching OSes. If I'm running a game, does it slow down because it's timesharing the CPU with another OS?

This

[Anonymous Coward]

would it be possible to do a linload type load of windows
in linux,protected mode in user space or kernel space?

then when done with windows, i can click linload that is set to
run in dosmode and go back to linux?

uh vm86() anybody?

[Anonymous Coward]

whats wrong with just using dosemu, it can run elks, and lots
of doses

What mode does the guest OS run in?

[Anonymous Coward]

How *do* they do this? The usual (S/370, 680x0) method) is to run the guest O/S in user state. The VM supervisor thus intercepts any access to privileged resources and provides a virtual copy. When an OS wants to verify that it is running privileged, for example, it reads the status register (CPL, in Intel terms). This read generates a trap to the host O/S, which stuffs the correct virtual privilege level into the guest's context and returns.

But this doesn't work in a Pentium, because most privileged registers are accessible in user mode. That is, no context is generated, and a guest O/S will get the "wrong" value. I suppose you could alter the O/S to intercept those reads if you knew where they were: in NT, where there is a defined layer between machine-independent and machine-dependent code, and in Linux, where you can find the reads in source code, if nothing else. But what do they do for a black-box O/S?

Ask(nicely) VMWare to rethink pricing

[Anonymous Coward]

First off, I think people should actually go
to VMWare's site and see what this is all about
before trying to spout off about what can and
can't be done. Secondly wait for the beta that
will be out in March and we can all see how well
it works.

Lastly at $300 this is quite pricey, but do what I did and write the sales department(there's an e-mail link right on their page) and ask them to
rethink their pricing model stating that at $300 dollars you probably wouldn't be buying it. If they get enough statements like this maybe they'll reduce the price significantly.

Consider that if this software works as advertised this could be one more piece in the puzzle to get Linux on the desktop. By allowing workers to run those Win95 apps they HAVE to have(read Office) and slowly migrating them to true Linux apps.

i was at demo

[Anonymous Coward]

and i talked to those guys - the person said first of all there is like a 30% performance hit. my next question was , naturally, waht about transmeta? and then she smiled and said 'we're intimately familiar with them' ... thats all i got but i got the feeling they are planning to leverage tm's stuff.

samedi@disinfo.ihatespam.net

Been There - Done That

[Anonymous Coward]

The IBM VM operating system has indeed been doing this for a long time. However doing this in software is the old ugly way. For over 10 years (13?) Amdahl and IBM mainframes have used hardware assisted machine partitioning. Amdahl calls it Multiple Domains Facilities (MDF) and IBM calls if Logical Partitions (LPAR). I think both methods work by partitioning up the memory into dedicated blocks and including extra hardware to do very fast state switches (store and load all registers). On machines with multiple CPUs, the CPUs can be shared (timesliced) or dedicated to a specific OS. The overhead is very low. I think arround 2%. Alomost 0% with dedicated CPUs. I've been waiting for years for Intel to "invent" this for PCs.

vm386 Mode

[Anonymous Coward]

How hard is it to impliment a 386 Virtual Mode on the chip. If AMD impliments this, they can get a leg up on Intel.

my two cents

[Anonymous Coward]

This was actually IBM/Apple's original plan for the PowerPC line -- OS/2, AIX, MacOS, and Taligent all running on a single microkernel. Rather than emulating the legacy OSs, you could just switch back and forth.

It also sounds like what Apple is up to with MacOS-X. Ye olde MacOS and NextStep running on the same microkernel. You wouldn't think they could do this with MacOS, but they actually have been for a long time with the old A/UX Unix.

"This is possible because each VM has ..."

[Anonymous Coward]

>Win95 inside OS/2 nearly impossible - they each want to control the basic chip.

Maybe someone who really knows OS/2 may want to comment on this. My understanding was that OS/2 could do 386 virtualization - i.e it could run 386Enhanced Windows programs and some Win32s stuff. Supposedly, the decision not to do a Win95-OS2 was for marketing reasons, not technical ones.

There was also that version of OS/2 that ran a full version of NetWare 3.x in a VM box. I would think that NetWare is all 32-bit code.

AMD/ALPHA/Cyrix.

[Anonymous Coward]

I think the Alpha does have support for something similar (thus "Galaxy", which was mentioned). Remember, it's not at all related to x86, though.

Overscan

[Anonymous Coward]

Now don't forget the Commodore 64 with its border sprites.

economics

[Anonymous Coward]

Linux + VMWare + WinNT OEM packs + VNC = WindowsNT Terminal Server - $2500 + interesting legal questions.

1. Build a BEEFY Box with 4 ethernet cards and much RAM, much drive.
2. Install Linux.
3. Install VMWare.
4. Install NT and other OSes over VMWare
5. Install VNC on NT and other OSes. (look it up if you don't know)
6. Run MULTIPLE INSTANCES of NT running VNC, so your users can have a virtual computer accessable anywhere with IP.
7.Stop scratching your head over what your Microsoft liscenses actually mean.
8. Call your legal department and have them figure it out.
9. Begin migrating to Applixware StarOffice or other unix based productivity software made by reputable developers before Microsoft gets wise and cooks up some other sort of inquisition for you.

Is this new?

[Anonymous Coward]

Univac's Exec8 operating system back in the '70s had a debugger that allowed you boot up operating systems inside it and run them just like any old user program you were debugging. Great for hacking the kernel and testing fixes, wonderful for trying out new OS releases, all while the rest of user land ran on oblivious and unaffected.

Oh, and Exec8 came with source (as I think all O/Ss did back in that golden age). It was of course derigeur back then to run your own (heavily) hacked system. There was a huge and active traders market in freeware utilities and patches. As they say, everything old is new again.

re: You write like a Fucking Geek(tm)

[Anonymous Coward]

Why can't you semiliterate slashdot geeks just write in English?? This trope is almost as bad as that idiotic *grin* *grin* Taco-head sticks everywhere. What a bunch of morons!

This has been another Troll sponsored by Prozac(tm)! Remember, when you think 'mentally unballanced', think Prozac(tm)! Thank you for watching.

Been There - Done That

[Anonymous Coward]

Got the t-shirt. Trap all the priv ops and emulate them. Use virtual memory to fake out the virtual machine. Figure out how to virtualize all the shared hardware resources. It's fun once you realize the what hardware architecture is really saying is that you can cheat as much you want as long as you don't get caught.

An virtual machine can be useful. Just think of those operating systems that don't really multitask too well. Can't think of too many of those can we. Now you can just multitask them at the virtual machine level. Anyway would you waste a entire real machine on Windows 98?

Um, Coward: 0, Taco: 1

[Anonymous Coward]

Ahem, let me explain how it works.

The A20 line is toggled via an I/O device (the KBC). You can virtualize I/O and memory in protected mode. The real challenge which they had was that some non-priveledged instructions give away that a piece of code is running as a user process. The solution: pre-scan the code, and replace these instructions (smsw, pushf, popf, and a few others) with invalid instructions. These are then trapped, and emulated, to provide the full effect of being alone on the machine. Self modifying code complicates things-- perhaps periodical scans of the code are necessary.

How they do it?

[Anonymous Coward]

Wait... doesn't VM86 impose some other real mode-esque limitations, such as 64k segments?

If anyone is looking for a cool hack...

[Anonymous Coward]

If someone ever wants to do a really cool and useful hack, here's something that I thing would be feasible, but hard. Make Linux and Windows 95/98 run together at the same time on dual (or more) CPU systems.

You'd have to make a Linux that would stay off the main CPU, and you'd have to give Windows ownership of the display and the disks and other hardware. You'd write Linux display and disk drivers that really communicate with Windows VxDs on the main processor to access the devices.

This main problem is that to pull this off, you'd have to be a pretty good Linux kernel hacker and you'd have to be a pretty good Windows VxD hacker, too. I don't think there is a lot of overlap between those groups.

--Tim Smith

DOSEMU Complete virtual machine

[Anonymous Coward]

(I'm not an Anonymous Coward-- I just don't remember my password-- Warren E. Downs, vwdowns@bigfoot.com)

I have been interested in doing a complete x86 virtual machine implemenatation for some time. I know that
the x86 CPUs don't fully virtualize themselves, so it would seem
impossible (or at least very hard) to get some programs or operating
systems to run under DOSEMU.

However, after hearing of this company that claims
to be doing just that, I thought about it a little. It seems it would be possible to virtualize the CPU. Here is my idea:

* Scan a portion of memory that is about to be executed, looking
for any bytes that *could* be part of a non-virtualizable instruction
(for instance, as I understand it, the instruction to obtain the
current CPU protection ring doesn't generate a trap, so it is
impossible for an emulator to tell the program that it is indeed
running in ring zero, when it is actually running in ring 3).

In this case, we would detect the byte(s) that would identify
that instruction to us. Note that we could also locate other
instructions that happen to contain the same byte value at a
different point in the instruction.

* Set a hardware breakpoint on each of the bytes we locate
(how many breakpoints can be set simultaneously?). This
would require a higher level CPU (pentium?)

* Start execution.

* Whenever we hit a breakpoint, we check to see if this
is actually a non-virtualizable instruction, and if so,
emulate the functionality we need (e.g. set the flags to
make the running program think it is at ring zero).

* Continue execution.

This idea may need to be tuned for specific details I have
missed.

I don't know if any of you are interested
in making something like this work, but if you or anyone
you know is interested, I'd love to hear from you. Please bring this to the attention of anyone you know who has technical knowlege of the x86 CPUs and would be interested in such a project.

Does this idea sound feasible? What would it take to
implement it as part of DOSEMU?

Warren E. Downs
vwdowns@bigfoot.com
http://wdowns.businesslink.com

$300...

[Anonymous Coward]

Check out www.freedows.net
A GNU OS doing the same thing..

SGI has similar plans for Irix?

[palpatine]

Why don't they just include the necessary libraries to run Linux/MIPS applications on IRIX/MIPS? Like glibc, etc.

Where's the source?

[sterwill]

Sounds like a neat toy to have, if I ever rebooted for anything.

Interesting

[Mark Pitman]

When I first was reading the article, I thought you would install this VMWare thing on your system first, then install other OS's on top of it. That seems like that would be a better way to go. I don't know a whole lot about this type of stuff, but it just seems like that would be faster?

Is this new?

[Matts]

Hi Ben!!!

It's probably not new - but let's watch them be granted a patent on it anyway.... :-)
--

Details lacking

[krynos]

The 386 and later processors had almost all special functions privilegied (memory table modifications and hardware access), except the function that enable to check in which ring (0 is kernel level, 3 is user level, some OS use level 1 and 2). The problem is the OS (Linux, NT, Win95) or DOS drivers (QEMM, EMM386, ...) will check in which level you are and refuste to run if you are in V86 mode or rnig other than 0... Except the instruction to query the processor ring level, the others important ones are privilegied and can be emulated.

With lots of clever tricks you could intercept this instruction, that's what Merge (under SCO and others) and probably VMware does.

I could imagine a way to intercept this instruction that would not completly kill the performance (in real mode the performance it would be bad, but not fatal, in protected mode, almost nil)... e-mail me for further details...

Interacting with hardware?

[Eric Sharkey]

I have a hard time imagining how interacting with certain pieces of hardware would work. I mean, if we just consider a PNP sound card, which OS initializes it? When the card generates an interrupt, which OS handles it?

With the price of computers coming down as quickly as they are, I think I'd rather just buy two computers if I really need to run two operating systems concurrently. The cost of a basic headless PC isn't that much more than the $300 asking price for the software and you don't have all of these kinds of headaches. Sounds like a bad idea to me.

Hmm... Better.

[red_dragon]

At least this VMware stuff sounds better than that big load of bollocks about the reconfigurable supercomputer we had yesterday. It even seems quite plausible (remember that you can achieve a similar effect with OS/2, which can load another operating system while it is running, although only those that don't attempt to run in ring 0, like DOS, CP/M-86, Xenix, and the Win95 setup disk). We'd still have to see how this how this software would manage sharing of hardware resources (how can both OS's talk to the SCSI controller or Ethernet card at the same time?), and how it would compare to hardware partitioning a la IBM ES/9000.

xfree can do that.

[mattdm]

No problem. You can configure XFree86 to use whatever weird resolution your card/monitor can deal with.

--

Hey Cmdr I gave a link to MySpace over a Week ago!

[narrowhouse]

Am I missing something? I submitted a link to Pixel Co. over a week ago and it never got posted. Around the same time I posted a reply to an article that never showed up. Is it because of the database problems? My breath? I'll brush, really!

Is this new? - DEC has done this too

[Jeff DeMaagd]

DEC had (has?) a system called Galaxy that is capable of running WinNT, Digital Unix and OpenVMS on a 6 CPU system, and CPUs can be exchanged between operating systems.

I really just want an 8400: 144 PCI slots (who knows how many independent busses this is?), to 28GB RAM and 14 - 600MHz CPUs. Fun....

VirtualPC/RealPC for the Mac emulates PC hardware

[timur]

It is possible to get the multiple OS's to share the same hardware. Concurrency is somewhat of a problem, though. The solution is to completely emulate the hardware, which is exactly what VirtualPC and RealPC do on the Mac. They emulate a SoundBlaster Pro, and IDE controller, and S3 card, etc. Then they translate all I/O operations to the equivalent Mac API.

--
Timur "too sexy for my code" Tabi, timur@tabi.org, http://www.tabi.org

People don't try Linux because they don't need it

[timur]

Most people don't try Linux not because they're afraid to leave Windows, but because they don't need Linux.


--
Timur "too sexy for my code" Tabi, timur@tabi.org, http://www.tabi.org

my bad

[Binary Boy]

I shoulda checked the Sheepshaver site before I posted, theyve got status on their LinuxPPC port of Sheepshaver:

Status of the LinuxPPC version:

* PPC/68k run-time environment: partly done (interrupts only handled by 68k)
* OS utilities: done
* Time Manager: done
* Mouse: not yet implemented
* Keyboard: not yet implemented
* Video: under development
* Sound: not yet implemented
* Floppy: done except for disk change detection
* Disks: done except for disk change detection
* SCSI: no suitable API in Linux
* CD-ROM: done
* Host filesystem access: done (no resource forks)
* Ethernet: not yet implemented
* Serial I/O: done

Im surpised at the disparity with the BeOS project, I imagine they werent doing much via a vm and instead did a lot of the original version as Beos drivers. In any case if they can get a version done providing low overhead and stability there will assuredly be Mac users migrating to Linux in droves, if for no other reason than to follow the trend.

Oh, and multithreaded for SMP Id hope, though I seem to remember Sheepshaver running in a single thread on Be, which was also a dissappoint me as I have an SMP card which screams on Be.

my two cents

[tgd]

The x86 line can't run virtual 386+ machines in hardware. If their product really did what they say it does, then ANY os would work on it, not just WinNT and Linux.

My bets on them having recoded the lowest layer of the Linux kernel to abstract it a step from the hardware... Three years ago when I was bored out of my mind at work and was being forced to use NT, I'd thought about the possibility of having a core microkernel sitting directly on the hardware, and having customized kernels for Linux and NT that mapped their memory management, task switching, etc all through the microkernel.

I wonder if they're claiming a patent on it... I know I've got a whole notebook of notes sitting around on the idea. :)

machines aren't expensive anymore

[ThwartedEfforts]

It seems strange to work on something, or even use something, like this considering that the price of machines, as recently reported here on slashdot, is approaching zero.

Cool Concept

[Waxman]

Very cool idea, and I really like the fact that a linux version will be out be for the winnt :)

Anyway, I wonder if this will allow you to bind an os to a particular cpu in an smp type system. Now, that would really rock.

I already boot 3 OS`s Simultaneously

[SteveSgt]

On a typical work day, I'll have 3 operating systems booted simultaneously on my Mac G3: The MacOS, SoftWindows95, and MachTen (BSD 4.4).

This seems to be an example of someone else trying to catch up to the same idea.

I already boot 3 OS`s Simultaneously

[The Metahacker]

And I could be booting four -- just run MacOS under BeOS via SheepShaver...

I don't see why this is a big new idea.

Um, Coward: 0, Taco: 1

[warmcat]

From http://www.vmware.com/products/forwindowsnt.html

For example, with VMware Virtual Platform, you can take your personal or business applications with you, and run them on the machine of your choice. Or you can run Windows NT concurrently with other operating systems such as Linux® on a single standard PC.

Bet you feel dumb now :) Never mind.

Is this new?

[Ben Hutchings]

Or is it just something IBM did 30 years ago. Hmm.

Flux/Fluke (Re: A bit steep)

[armb]

You might want to look at the University of Utah's
work on recursive virtual machines.
http://www.cs.utah.edu/projects/flux/
ftp://mancos.cs.utah.edu/papers/fluke-rvm-abs.ht ml

SGI has similar plans for Irix?

[beb]

I read somewhere that SGI was working towards dual-boot Irix/Linux boxes that don't actually require a reboot. Anyone know if this is similar technology?

If this VMware thing actually works, and the overhead is low (less than 10%) I'd buy it. I am a bit skeptical though...

-beb

Demo99

[GleekOid]

Things announced at Demo99 tend to be close to reality, i.e. not marketing BS. If this works, Redhat would be dumb not to buy into this company with some of their venture $$, just to be able to bundle it with their linux distro.

Somewhat myopic.

[hpa]

It's real. I saw it a couple of weeks ago; the VMWare people were kind enough to give a demo to a few of us Transmetans.

See the website

[Cato]

Have a look at their website, http://www.vmware.com/ - it's pretty informative, with architecture diagrams etc. Looks like a VM/390 style virtual machine hypervisor, i.e. each guest OS is transparently multitasked with other OSs. The only reason there are Linux and NT versions, IMO, is so that the control applications to toggle between OSs etc can run under an OS, and are therefore a bit easier to access. Any version should be able to run a wide range of guest OSs including Linux, NT, FreeBSD, Win98, etc.

I've put my name down for the beta - it's pricey but then the market is probably quite limited, and it's not trivial to develop.

Details lacking

[Ramana]

I have a hard time believing that something like
this can be done without modifying any of the
linux source code and NT source code. Well, I can
believe they tweaked Linux kernel here and there
but what about NT?

It would be even more bizarre if they say that we can take an off the shelf Redhat CD and NT CD and install on top of VMware.

I don't get it

[NYC]

The Linux version is available soon, but the NT version is not? How is this supposed to be running multiple-OSes. Plus, how many users have extra space around their desktop? I have a 17" monitor and no extra space. Besides, 2 OSes is not that enticing. I also have BeOS crammed onto my computer.

A bit steep

[Cerb]

$299 is a bit much for us OpenSource freaks. :) Maybe a WINE type project may arise from this.

No Subject Given

[Cerb]

Funny, but not quite what I meant. I was going for the "reverse engineer" or re-implementation of APIs and such. So I guess SAMBA might fit too?

By the way, I'm sorry you think WINE sucks... Works for what I "need". :)

Could be a real Windows killer/Linux eye opener.

[orignal]

This is exactly what we need. Keep the possibility to run Windows so it is not painful to try out Linux.

Most people don't want to switch to Linux because that are scared of not having all their little WinGizmos anymore.

Slowly, people will stop using Windows and start using Linux more and more... After a while, Windows will only be used for games, its true nature since it is a toy anyway.

Concurrent OSes

[K.]

Do a search for (Compaq OR Digital) AND Galaxy.
Schweet.

K.
-

my two cents

[Col. Klink (retired)]

> If their product really did what they say it does, then ANY os would work on it, not just WinNT and Linux.

So you didn't read the web page either? Linux or NT are just the HOST NT. With the Linux version, you can indeed run ANY OS.

> I wonder if they're claiming a patent on it... I know I've got a whole notebook of notes sitting around on the idea. :)

As, I'm sure, does IBM. Perhaps something about their IMPLEMENTATION is patentable, but not the notion of Virtual Machines.

"This is possible because each VM has ..."

[wapentake]

Virtualization of the 386 is pretty simple. Only the privileged instructions require emulation. Non-privileged instructions execute directly on the processor. The vmware executes NT and Linux as processes with normal, user-level security access. Thus when they attempt to execute priveleged-level code, they trap into vmware, which massages the instructions. For example, vmware will catch a TLB insert, and modify the mapping to match that which vmware desires.

The designers of Disco (specifically Rosenblum) are behind vmware, so you can understand the product by reading the Disco papers at Stanford. Look for the FLASH project.

Disco implements the vmware concept, but for IRIX. Their papers describe the performance impact caused by the virtualization, which is fairly negligible. If your program executes mainly non-privileged instructions, then the virtualized machine has no impact. Boot-up time of the OS probably sees the greatest impact, due to processor initialization, address space construction, and hardware probing.

The performance impact caused by virtualized hardware devices can be minimized by coding drivers for NT and Linux that interact directly with vmware. I suspect vmware followed that route.

Oh great

[Maniacal]

As if all the crap I have running at one time isn't enough to chew up my resources. Now I can add another OS to the pile. I'll bet the RAM and processor manufacturers are loving this.

M

Awesome

[joshv]

This looks like the Real Thing(tm).

From what I can tell from their web site, you boot a primary OS, NT or Linux are the only two they going to support so far - and then from that OS you launch virtual machines that should be able to run any OS.

I guess this is similar to how NT runs 16 bit windows apps.

The only thing is, I would have though that the x86 chip would have to be able to virtualize itself entirely, i.e. provide a virtual machine with all the same capabilities of the actual chip itself. From what I understand, the VM mode of the x86 chips is a limited subset of the features of the chip itself.

Also performance could be weird. It looks like each VM is just a process in the host OS, subject to the scheduling of the host OS. If you boot Linux from NT and NT decides to give a native process 90% of CPU, it doesn't look like there is much Linux, or the Virtual machine manager can do about it.

-josh

uh vm86() anybody?

[dirty]

actually it is new. The vm86() call cannot, and never will allow you to run another 32bit os. It supports DOS great because DOS doesn't put the cpu in "protected mode". Intel CPUs only have hardware support for virtualizing 8086s, an 80386 cannot be virtualized in hardware on an intel chip. Too bad they never bothered to fix this.

How do you switch? Wheres the BeOS or DOS version?

[dirty]

Then just setup a small linux partition which does virtually nothing from after the kernel boots. The kernel takes very little time to load, it's mostly the init scripts that take a while.

Calm down.

[dirty]

Why do people always feel the need to be so violent? Take a deep breath, drink some water.

Some Comments

[jerodd]

Firstly, let me say that PixelWare appears to be very sour vaporware. I called them last year when I saw something about their product in InformationWeek. After pressing the question, they said ``yeah, it runs in Windows 95. Unix support will be out soon.'' That was six months ago. Read: it's not happening.

Virtual machine emulation is interesting, however. Freed software like GNU/Linux or opensource software like BSD is, quite obviously, the easiest to encapsulate on a new platform. I've already got Linux (kernel 2.2.1 <grin>) sort of working in OS/2; I junked things like the memory-manager, and created a special filesystem that just talks to OS/2, rather than actually accessing disks. It is absolutely not complete: things like networking are missing and direct hardware access does not work yet. I really want to get in touch with the Linux kernel people and work on a new architecture--perhaps a VM architecture, for running on other platforms.

I've also toyed with running Windows under other operating systems (in my case, it's Windows 95 under OS/2, but the same would apply for any OS under something like GNU/Linux or BSD). Here's what I've done so far:

I started with Bochs, and put it in tracing mode. Then I booted Windows 95 up to the point where it loads the GUI. I took the resultant 750MB trace file (it took hours to generate) and began analyzing it for things like privileged instructions. Specifically, I wanted to watch the call flow for calls to the memory manager.

Next, I tried hooking the memory manager calls (this was quite simple) and making some changes to Bochs

This stuff is nowhere near pre-alpha quality yet. I'm more actively working on XCLIENT for OS/2 and some other OS/2 software at the moment. However, if you're working on a VM and would like to chat, feel free to drop me a note. It's not as hard as you think, and I did get Windows 95 to boot with my special memory manager.

I also tried recoding Bochs to execute instructions natively, but that did not work. What looks like it might be promising is hooking the memory manager so that instructions that access memory (such as MOV EDX,[EAX]) can run natively. I haven't tested this, though. Emulating I/O hardware is silly, however--just write Windows 95 device drivers that place calls to the host operating system.

Cheers,
Joshua.
You can see --jon. Postel here. Love that beard!

Sheepshaver

[krb]

At the risk of bringing up both the MacOS & BeOS in such a Linuxcentric realm, may I mention Sheepshaver. It's a $50 prog for BeOS which allows the MacOS to run in a window (or fullscreen) at native speed, concurently with the BeOS, allowing data transfer between os's, etc. A version for LinuxPPC is underdevelopment. The sheepshaver site is http://www.sheepshaver.com, for any interested folks.

-k-

More on extra screen space -- The Pixel Company

[takshaka]

http://www.thepixelcompany.com/myspace/product/ind ex.html

"It's not Windows-dependent - which makes it possible to run two operating systems (like Windows and Unix) independently, and simultaneously."

The system requirements, however, say only Win95/98.

"...the MySpace control bar has limitless content possibilities. It can launch any technology - it can manage applications; manage hardware; link to the Internet; control your CD; WebTV, and more. It can also be used to add Java applets and other operating systems. But most importantly, MySpace is always right in front of the user."

It sounds like a FreePC sort of advertising area drawn outside the standard resolution scan area.

-tak

SMP

[Darmox]

I had a bit of a thought about this... Say you are running the Linux version. That'll be one process, right? Under that you run NT or 95... Whichever, it's not important. Anyways, 95 for instance can't use multiple processors. BUT, Linux can. So, if the app is threaded, could you expect the increase in performance? Even if not, you could still have one processor for Linux and one doing 95 stuff... Anyways, looks like some really cool stuff... Just a thought on it.
Darmox

um, What is '"extra" screen space?'

[Mad Hatter]

I don't know about anyone else but I have never had any "extra" screen space, as a matter of fact, I never seem to have enough.

"Trouble is, just because it's obvious doesn't mean it's true"

GOD --- LEARN TO READ!!

[kijiki]

maybe you should take your own advice, the web page clearly and repeatedly states that it runs real mode or protected mode OSes, including NT4. This of course is not new, since you can do this on SCO unix using "Merge."

You write like a Fucking Geek(tm)

[linuchristo]

You are angry that you don't know what a self-virtualizing instruction set is; huh, guy?

Overhead?

[eon]

How will running two O/S's affect users like myself running lowly p166's? No Dice.

Dichotomy

[Anonymous Shepherd]

So at a certain level, before mainstream acceptance, programs like this will be seen as a try before you buy type scenario...
"I've heard all these neat things about Linux, but I still need to use WinNT for my daily operations..."
This user then has the chance to try out Linux as painlessly as possible, if VMWare isn't a hoax, because Linux will just be another program on an NT machine. The user will then come to realize that Linux is stable and can do most of what is needed, and switch accordingly.

However, a dearth of applications will prevent people from trying, not just using, Linux, so this program fulfills some of that loss. VMWare is going to help Linux just through sheer exposure.

The problem I see is that if Linux takes more resources while under NT because of overhead, it's much vaunted stability and resource thriftiness may be lost. We shall see.

AS =)

SMP implementation

[Tekmage]

Now, if this thing will work on a dual-processor machine with a dual-headed video card - i.e. one proc + one head per OS - then I'll definitely be interested...

Have to keep an eye on this one.

What about Bochs???

[Scott Dudley]

I've not heard anyone mention bochs written by Kevin Lawton. I found out about it here, on Slashdot. It's a virtual 386 that will run on any Unix flavor (Mac and Win too). I created a 400MB disk image, fdisked it, and installed Windows 95. Was cool to see Windows running in a Linux Xterm. Too slow for everyday use but he's working on some tricks to speed things up.

wow man

[Llewyn]

keep me posted on that!

"This is possible because each VM has ..."

[Vidar Hokstad]

I remember being flamed over and over a few months back for claiming that this was possible :-)

Intel CPU's can't be fully virtualized. However, there's nothing stopping someone from virtualizing most of it, and "verify" code before you let it run natively. Most code would never do anything that would give you any problems, and could be executed directly. The small parts that would cause problems, can be modified to call a handler that emulate whatever the original code tried to do.

Of course this causes code to run slower, as you'd have to do a lot of work to ensure the integrity of the system, but it would still be a lot faster than a full emulater like Bochs.

a great idea

[ash]

ROFL