Re: [vbox-users] Parallel machines
Pablo Sanchez wrote: > On Friday 12 December 2008 at 12:24 pm, Peter Ondru?ka penned > about "Re: [vbox-users] Parallel machines" > > >> Speaking of Christmas reminds me to wish you Merry Christmas. >> > > Yes, happy holidays to the entire VirtualBox team. > > Cheers, > Yes Happy Christmas to VitualBox team and Sun for support for opensource projects. ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Fri, Dec 12, 2008 at 2:24 PM, Peter Ondruška wrote: > Speaking of Christmas reminds me to wish you Merry Christmas. And many > thanks for bringing Virtualbox on Mac. This version 2.0 is really > good--I have replaced Parallels 3 (ie no more upgrade of > Parallels...). Peter Good! Maybe one day even Mac users will see the Light and the true benefits of Open Source software. Speaking of Sun, do you know StarOffice for OS X is now available, don't you? :) (sorry, couldn't resist ... I'm a bit tired of seeing Apple get all the press and Ubuntu and FOSS get ignored ;-) FC ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Friday 12 December 2008 at 12:24 pm, Peter Ondru?ka penned about "Re: [vbox-users] Parallel machines" > Speaking of Christmas reminds me to wish you Merry Christmas. Yes, happy holidays to the entire VirtualBox team. Cheers, -- Pablo Sanchez - Blueoak Database Engineering, Inc Ph:819.459.1926 Fax: 760.860.5225 (US) ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
Speaking of Christmas reminds me to wish you Merry Christmas. And many thanks for bringing Virtualbox on Mac. This version 2.0 is really good--I have replaced Parallels 3 (ie no more upgrade of Parallels...). Peter 2008/12/12, Frank Mehnert : > On Friday 12 December 2008, Maciek Kaliszewski wrote: >> Are there any plans to "emulate" multicore processor inside VBox ? >> I mean if host computer has multiple cores - single VBox guest instance >> will see ( and will be able to run multiple processes on them ) some of >> the cores not one (correct me if I'm wrong ) as today. > > Yes, of course! We are working on this but it will still take some more time > (probably months). Sorry, not until Christmas anymore ... ;-) > > Kind regards, > > Frank > -- > Dr.-Ing. Frank MehnertSun Microsystemshttp://www.sun.com/ > ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Friday 12 December 2008, Maciek Kaliszewski wrote: > Are there any plans to "emulate" multicore processor inside VBox ? > I mean if host computer has multiple cores - single VBox guest instance > will see ( and will be able to run multiple processes on them ) some of > the cores not one (correct me if I'm wrong ) as today. Yes, of course! We are working on this but it will still take some more time (probably months). Sorry, not until Christmas anymore ... ;-) Kind regards, Frank -- Dr.-Ing. Frank MehnertSun Microsystemshttp://www.sun.com/ signature.asc Description: This is a digitally signed message part. ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Friday 12 December 2008, Marc Beck wrote: > We are talking here about VirtualBox, not about a hypervisor, so if you are > correct, I would be glad to see some documentation about this. This has nothing to do with being a hypervisor or not. Just have a look at the source code, it is freely available. Or just open two VMs on your favorite OS and check how many VirtualBox processes you see. Kind regards, Frank -- Dr.-Ing. Frank MehnertSun Microsystemshttp://www.sun.com/ signature.asc Description: This is a digitally signed message part. ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
Sorry if I offended anybody, or sounded as a know-it-all, I am sure Frank knows ways more than I do. Just trying to understand. From the question I take we are talking of a virtual box, running on top of a Host. As I was told, no matter what, the host is in-between the virtual machines and the bare metal. Marc 972-800-2150 www.Amazingcomputing.biz -Original Message- From: vbox-users-boun...@virtualbox.org [mailto:vbox-users-boun...@virtualbox.org] On Behalf Of Stealth Sent: Friday, December 12, 2008 10:18 AM To: VirtualBox end user list Subject: Re: [vbox-users] Parallel machines On Friday 12 December 2008 11:06:30 am Marc Beck wrote: > We are talking here about VirtualBox, not about a hypervisor, so > if you are correct, I would be glad to see some documentation > about this. Interesting that you would disagree with some one from Sun Microsystems. Frank is probably one of the developers of VirtualBox. If so, he would know how things work better than you. -- Stealth ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
Frank Mehnert wrote: > Marc, > > On Friday 12 December 2008, Marc Beck wrote: > >> I doubt that even with multiple cores it will make any improvements. >> That's because the way the multicore OS is designed. Basically, as far as >> I know, the load is not shared in such a way that one application would run >> on one pc, and a second one on the other. On top of that, from the host >> point of view, the VM shell is a single application, no matter how many VM >> you get to run within that shell. Unless there is a way to designate a >> specific core for a given VM within a single shell, its a no go. >> > > That is wrong. Your understanding of how a VMM works is not correct. There > is no VM shell. Each VM has its own process. Therefore it _will_ make a > difference if you compute in two or more VMs in parallel on a multicore > system. Apart from this, even a VM process uses several threads which can > run in parallel if they don't wait for each other. So even a single VM > is faster on a multicore system than on a single core system if you are > able to utilize the threads. For instance when doing I/O: One thread is > emulating the guest, another thread is doing the work on the host. > > Are there any plans to "emulate" multicore processor inside VBox ? I mean if host computer has multiple cores - single VBox guest instance will see ( and will be able to run multiple processes on them ) some of the cores not one (correct me if I'm wrong ) as today. Best regards Maciek Kaliszewski ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Fri, Dec 12, 2008 at 10:06 AM, Marc Beck wrote: > We are talking here about VirtualBox, not about a hypervisor, so if you are > correct, I would be glad to see some documentation about this. > > Thanks > > Marc > 972-800-2150 > www.Amazingcomputing.biz > > Hi, VirtualBox is a hypervisor. And it's one that runs either beside or on top of a HostOS. Thus you're going to slow down because of the abstractions needed to isolate resources for each VM. The problem with using multiple VMs to compute a problem with say MPI is that you are still using _one_ machine. So it's one machine's resources being used the entire time. With one machine's processor (multi-core or not) having to swap out the context of both the HostOS AND any VMs that are currently executing. Virtualization takes too big of a hit due to abstraction to get a speed increase. Especially using the kind of hypervisor VirtualBox is. No offense to the virtualbox devel team. it's a great hypervisor for what it's trying to accomplish and that's not really this. Thanks, Ferrol ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Friday 12 December 2008 11:06:30 am Marc Beck wrote: > We are talking here about VirtualBox, not about a hypervisor, so > if you are correct, I would be glad to see some documentation > about this. Interesting that you would disagree with some one from Sun Microsystems. Frank is probably one of the developers of VirtualBox. If so, he would know how things work better than you. -- Stealth ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
We are talking here about VirtualBox, not about a hypervisor, so if you are correct, I would be glad to see some documentation about this. Thanks Marc 972-800-2150 www.Amazingcomputing.biz -Original Message- From: vbox-users-boun...@virtualbox.org [mailto:vbox-users-boun...@virtualbox.org] On Behalf Of Frank Mehnert Sent: Friday, December 12, 2008 9:58 AM To: vbox-users@virtualbox.org Subject: Re: [vbox-users] Parallel machines Marc, On Friday 12 December 2008, Marc Beck wrote: > I doubt that even with multiple cores it will make any improvements. > That's because the way the multicore OS is designed. Basically, as far as > I know, the load is not shared in such a way that one application would run > on one pc, and a second one on the other. On top of that, from the host > point of view, the VM shell is a single application, no matter how many VM > you get to run within that shell. Unless there is a way to designate a > specific core for a given VM within a single shell, its a no go. That is wrong. Your understanding of how a VMM works is not correct. There is no VM shell. Each VM has its own process. Therefore it _will_ make a difference if you compute in two or more VMs in parallel on a multicore system. Apart from this, even a VM process uses several threads which can run in parallel if they don't wait for each other. So even a single VM is faster on a multicore system than on a single core system if you are able to utilize the threads. For instance when doing I/O: One thread is emulating the guest, another thread is doing the work on the host. Kind regards, Frank -- Dr.-Ing. Frank MehnertSun Microsystemshttp://www.sun.com/ ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
Marc, On Friday 12 December 2008, Marc Beck wrote: > I doubt that even with multiple cores it will make any improvements. > That's because the way the multicore OS is designed. Basically, as far as > I know, the load is not shared in such a way that one application would run > on one pc, and a second one on the other. On top of that, from the host > point of view, the VM shell is a single application, no matter how many VM > you get to run within that shell. Unless there is a way to designate a > specific core for a given VM within a single shell, its a no go. That is wrong. Your understanding of how a VMM works is not correct. There is no VM shell. Each VM has its own process. Therefore it _will_ make a difference if you compute in two or more VMs in parallel on a multicore system. Apart from this, even a VM process uses several threads which can run in parallel if they don't wait for each other. So even a single VM is faster on a multicore system than on a single core system if you are able to utilize the threads. For instance when doing I/O: One thread is emulating the guest, another thread is doing the work on the host. Kind regards, Frank -- Dr.-Ing. Frank MehnertSun Microsystemshttp://www.sun.com/ signature.asc Description: This is a digitally signed message part. ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
I doubt that even with multiple cores it will make any improvements. That's because the way the multicore OS is designed. Basically, as far as I know, the load is not shared in such a way that one application would run on one pc, and a second one on the other. On top of that, from the host point of view, the VM shell is a single application, no matter how many VM you get to run within that shell. Unless there is a way to designate a specific core for a given VM within a single shell, its a no go. Marc 972-800-2150 www.Amazingcomputing.biz -Original Message- From: vbox-users-boun...@virtualbox.org [mailto:vbox-users-boun...@virtualbox.org] On Behalf Of Mikkel L. Ellertson Sent: Friday, December 12, 2008 8:50 AM To: VirtualBox end user list Subject: Re: [vbox-users] Parallel machines Stealth wrote: > On Friday 12 December 2008 09:06:30 am Ondrej Sluciak wrote: >> I have one very stupid question, but I want to be sure that I am >> right. If you have 2 interconnected virtual machines running and >> let them compute some problem in parallel (using message passing >> interface - MPI), is it really a parallel computation? I mean, >> when the host machine has only one core and both VMs use the same >> memory and resources (though separated), can you talk about >> parallel computation? >> I'm asking because I did some tests with with such a setup, but I >> have noticed no improvement in speed, comparing to computation on >> one VM. Does anyone have some experience with it? Thanks. > > If you mean are you getting quicker results because you are using > multiple machines? The answer would be, no, because you are using > one machine. You are simply sharing the one machine's resources > multiple applications. The VMs are just a process running inside an > application. Each VM thinks it is a separate machine and the host > machine just simply sees an application demanding resources. > If anything, I would expect slower results using 2 virtual machines because of the added overhead of the second virtual machine. I am not sure even having more then one processor or core would help, because I suspect the VirtualBox is not designed for multiprocessors or cores. One way to find out would be to open more then one vm and watch the cpu load. I do not have a second vm set up right now to test it. I do know that running with one vm, I only use one core. This is handy because it automatically limits VB's cpu usage. Mikkel -- Registered Linux User #16148 (http://counter.li.org/) ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
Thank you for your answers. It is as I thought. I tested performance of 2 VMs on one single core host computer, but as Stealth said, there has been no increase in speed. I tried to parallely compute the multiplication of two huge matrices, but the time spent on computation was just the same as for the computation on one computer. I tried it with several sizes of matrices, but with there was no improvement even in the big dimensions (where you can assume, that the communication overhead is irrelevant in comparison to the computation time). Ondrej. Mikkel L. Ellertson wrote: > Stealth wrote: > >> On Friday 12 December 2008 09:06:30 am Ondrej Sluciak wrote: >> >>> I have one very stupid question, but I want to be sure that I am >>> right. If you have 2 interconnected virtual machines running and >>> let them compute some problem in parallel (using message passing >>> interface - MPI), is it really a parallel computation? I mean, >>> when the host machine has only one core and both VMs use the same >>> memory and resources (though separated), can you talk about >>> parallel computation? >>> I'm asking because I did some tests with with such a setup, but I >>> have noticed no improvement in speed, comparing to computation on >>> one VM. Does anyone have some experience with it? Thanks. >>> >> If you mean are you getting quicker results because you are using >> multiple machines? The answer would be, no, because you are using >> one machine. You are simply sharing the one machine's resources >> multiple applications. The VMs are just a process running inside an >> application. Each VM thinks it is a separate machine and the host >> machine just simply sees an application demanding resources. >> >> > If anything, I would expect slower results using 2 virtual machines > because of the added overhead of the second virtual machine. > > I am not sure even having more then one processor or core would > help, because I suspect the VirtualBox is not designed for > multiprocessors or cores. One way to find out would be to open more > then one vm and watch the cpu load. I do not have a second vm set up > right now to test it. I do know that running with one vm, I only use > one core. This is handy because it automatically limits VB's cpu usage. > > Mikkel > -- Dipl.-Ing. Ondrej Sluciak Room CG-04-06 Vienna University of Technology, Austria Institute of Communications and Radio-Frequency Engineering Gusshausstrasse 25-29/389 http://www.nt.tuwien.ac.at ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Friday 12 December 2008 09:38:46 am Heinz Wiesinger wrote: > Stealth: > > On Friday 12 December 2008 09:06:30 am Ondrej Sluciak wrote: > > > I have one very stupid question, but I want to be sure that I > > > am right. If you have 2 interconnected virtual machines > > > running and let them compute some problem in parallel (using > > > message passing interface - MPI), is it really a parallel > > > computation? I mean, when the host machine has only one core > > > and both VMs use the same memory and resources (though > > > separated), can you talk about parallel computation? > > > I'm asking because I did some tests with with such a setup, > > > but I have noticed no improvement in speed, comparing to > > > computation on one VM. Does anyone have some experience with > > > it? Thanks. > > > > If you mean are you getting quicker results because you are > > using multiple machines? The answer would be, no, because you > > are using one machine. You are simply sharing the one machine's > > resources multiple applications. The VMs are just a process > > running inside an application. Each VM thinks it is a separate > > machine and the host machine just simply sees an application > > demanding resources. > > Is this also true for multi-core systems? Is every VM running on > the same core, or has each VM it's own core (as long as there is > still an unused available)? > That way you could probably make use of your cpu's full power in > certain scenarios without the need for smp in the guest. > > Grs, > Heinz That would depend on how the host allocates its resources. When running virtual software of any kind, the host perspective sees just an application demanding resources. The host has no idea there is another OS running. The host does see the VM on the network but does not know the VM is hosted by the host. The host thinks the VM is just another machine on the network. -- Stealth ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
Stealth wrote: > On Friday 12 December 2008 09:06:30 am Ondrej Sluciak wrote: >> I have one very stupid question, but I want to be sure that I am >> right. If you have 2 interconnected virtual machines running and >> let them compute some problem in parallel (using message passing >> interface - MPI), is it really a parallel computation? I mean, >> when the host machine has only one core and both VMs use the same >> memory and resources (though separated), can you talk about >> parallel computation? >> I'm asking because I did some tests with with such a setup, but I >> have noticed no improvement in speed, comparing to computation on >> one VM. Does anyone have some experience with it? Thanks. > > If you mean are you getting quicker results because you are using > multiple machines? The answer would be, no, because you are using > one machine. You are simply sharing the one machine's resources > multiple applications. The VMs are just a process running inside an > application. Each VM thinks it is a separate machine and the host > machine just simply sees an application demanding resources. > If anything, I would expect slower results using 2 virtual machines because of the added overhead of the second virtual machine. I am not sure even having more then one processor or core would help, because I suspect the VirtualBox is not designed for multiprocessors or cores. One way to find out would be to open more then one vm and watch the cpu load. I do not have a second vm set up right now to test it. I do know that running with one vm, I only use one core. This is handy because it automatically limits VB's cpu usage. Mikkel -- Registered Linux User #16148 (http://counter.li.org/) ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
Stealth: > On Friday 12 December 2008 09:06:30 am Ondrej Sluciak wrote: > > I have one very stupid question, but I want to be sure that I am > > right. If you have 2 interconnected virtual machines running and > > let them compute some problem in parallel (using message passing > > interface - MPI), is it really a parallel computation? I mean, > > when the host machine has only one core and both VMs use the same > > memory and resources (though separated), can you talk about > > parallel computation? > > I'm asking because I did some tests with with such a setup, but I > > have noticed no improvement in speed, comparing to computation on > > one VM. Does anyone have some experience with it? Thanks. > > If you mean are you getting quicker results because you are using > multiple machines? The answer would be, no, because you are using > one machine. You are simply sharing the one machine's resources > multiple applications. The VMs are just a process running inside an > application. Each VM thinks it is a separate machine and the host > machine just simply sees an application demanding resources. Is this also true for multi-core systems? Is every VM running on the same core, or has each VM it's own core (as long as there is still an unused available)? That way you could probably make use of your cpu's full power in certain scenarios without the need for smp in the guest. Grs, Heinz signature.asc Description: This is a digitally signed message part. ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users
Re: [vbox-users] Parallel machines
On Friday 12 December 2008 09:06:30 am Ondrej Sluciak wrote: > I have one very stupid question, but I want to be sure that I am > right. If you have 2 interconnected virtual machines running and > let them compute some problem in parallel (using message passing > interface - MPI), is it really a parallel computation? I mean, > when the host machine has only one core and both VMs use the same > memory and resources (though separated), can you talk about > parallel computation? > I'm asking because I did some tests with with such a setup, but I > have noticed no improvement in speed, comparing to computation on > one VM. Does anyone have some experience with it? Thanks. If you mean are you getting quicker results because you are using multiple machines? The answer would be, no, because you are using one machine. You are simply sharing the one machine's resources multiple applications. The VMs are just a process running inside an application. Each VM thinks it is a separate machine and the host machine just simply sees an application demanding resources. -- Stealth ___ vbox-users mailing list vbox-users@virtualbox.org http://vbox.innotek.de/mailman/listinfo/vbox-users