Message boards :
Number crunching :
Crunching Virtually via VM
Message board moderation
Author | Message |
---|---|
Send message Joined: 2 Jan 09 Posts: 2 Credit: 93,156 RAC: 0 |
Just setup everything within a VM environment on one of my quad core boxes and it is running smoothly. Anyone else run projects on virtual machines (vm)? |
Send message Joined: 18 Nov 07 Posts: 4 Credit: 2,091,246 RAC: 0 |
Just setup everything within a VM environment on one of my quad core boxes and it is running smoothly. Anyone else run projects on virtual machines (vm)? I ran Cosmo for a while on my Apple Mac Pro using VMWare. It worked out nicely. I was running 2 dual core virtual machines which keep all 4 cores @ 100%. The overhead used by VMWare was surprising low. |
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
Just setup everything within a VM environment on one of my quad core boxes and it is running smoothly. Anyone else run projects on virtual machines (vm)? I've experimented with VMware on a core 2 duo. I'm not sure if the increased number of WU's running slower, and making everything else slower, is actually turning wu's around faster than just the 2 at a time. I'd be interested in you thoughts on whether VM can increase overall crunching. |
Send message Joined: 2 Jan 09 Posts: 2 Credit: 93,156 RAC: 0 |
Thanks for the input so far folks. I am currently running on a physical box with Vista Ult x64 with 8GB of RAM and have a single XP Pro x86 VM with 2GB of RAM running at the same time and all is stable after 12hrs or so. The physical box also has an nVidia 8800 GT video card which has GPUGRID and Cuda smoking it as GPU power is faster. All this within the same environment so it is interesting to watch. I want to also try and pull clients from physical side and make several VMs and run Milky Way, SETI (w/Cuda) and GPUGRID only and then try it all as a mixed bunch to see how results differ. VM will be setup as "barebones" OSs, both x64 and x86. I also have a Linux box running OpenSuse 10.3 and just SETI right now but plan of trying VMs in it for Milkyway as well as Cuda and GPUGRID for pulling GPU power. Also have another Quad core box here at the house if I need it. We'll have to wait and see as this will be a lengthy experiment. If anyone has done something similar already please share your results too. |
Send message Joined: 12 Apr 08 Posts: 621 Credit: 161,934,067 RAC: 0 |
I'd be interested in you thoughts on whether VM can increase overall crunching. It CAN, cut almost always won't ... Particularly on a computer with HT and large caches it is possible that by running more tasks than normal that you will actually use more of the HT capabilities of the system ... in other words, more threads in flight give a greater oppertunity to maximize the usage of the CPU resources. For example, my dual Xeon Mac Pro running VM ware with a dual processor in the windows box. I could be running BOINC on both OS with the Windows VM running as a dual core running two tasks ... we are at a total of ten tasks ... NOW, any time that the OS-X application, or the windows application blocks for OS resources or some other reason, there are other running tasks to take up and use the slack ... The down side is higher cache thrashing as things are pulled into the caches and forced out ... Fundamentally it is a question that only by testing can you prove if it is doing you good or not ... the problem is that almost no project is running tasks that are well predictable in run times ... and you would want to have a mix of projects to increase the chances of maximizing resource usage. In other words, running MW 10 times would likely see a decrease in throughput while running MW on the OS-X side and Cosmology or Prime Grid on the windows side MIGHT see improvements in THROUGHPUT, though not necessarily run times ... which is the other rub ... run time will almost certainly increase for all tasks running, but now you have 10 in flight instead of 8 and it might be worth while ... Another example is that you might want to increase throughput on a particular project that is not mac "friendly" and in that case you might use the VM side to run additional work even at the expense of slowing down production of the native OS-X applications ... like I want to run up the scores for Cosmology and Prime Grid as Windows only projects I might consider running two more though it will impact the running OS-X tasks I have now ... Of course, you can also try to dedicate one or more cores to the runing of the VM machine so that you are running the same number of tasks but are now running more of the other OS ... |
Send message Joined: 12 Apr 08 Posts: 621 Credit: 161,934,067 RAC: 0 |
I'd be interested in you thoughts on whether VM can increase overall crunching. It CAN, but almost always won't ... Particularly on a computer with HT and large caches it is possible that by running more tasks than normal that you will actually use more of the HT capabilities of the system ... in other words, more threads in flight give a greater oppertunity to maximize the usage of the CPU resources. For example, my dual Xeon Mac Pro running VM ware with a dual processor in the windows box. I could be running BOINC on both OS with the Windows VM running as a dual core running two tasks ... we are at a total of ten tasks ... NOW, any time that the OS-X application, or the windows application blocks for OS resources or some other reason, there are other running tasks to take up and use the slack ... The down side is higher cache thrashing as things are pulled into the caches and forced out ... Fundamentally it is a question that only by testing can you prove if it is doing you good or not ... the problem is that almost no project is running tasks that are well predictable in run times ... and you would want to have a mix of projects to increase the chances of maximizing resource usage. In other words, running MW 10 times would likely see a decrease in throughput while running MW on the OS-X side and Cosmology or Prime Grid on the windows side MIGHT see improvements in THROUGHPUT, though not necessarily run times ... which is the other rub ... run time will almost certainly increase for all tasks running, but now you have 10 in flight instead of 8 and it might be worth while ... Another example is that you might want to increase throughput on a particular project that is not mac "friendly" and in that case you might use the VM side to run additional work even at the expense of slowing down production of the native OS-X applications ... like I want to run up the scores for Cosmology and Prime Grid as Windows only projects I might consider running two more though it will impact the running OS-X tasks I have now ... Of course, you can also try to dedicate one or more cores to the runing of the VM machine so that you are running the same number of tasks but are now running more of the other OS ... |
©2024 Astroinformatics Group