Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
World Community Grid Forums
Category: Completed Research Forum: The Clean Energy Project - Phase 2 Forum Thread: Running CEP2 in a virtual machine. |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 19
|
Author |
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
An interesting discussion evolved over at the devs regarding a non checkpointing project having tasks that could easily take 11 days. What was recommended was to run these with BOINC in a Virtual Machine, which then took snapshots every so many minutes, which then if shut down and started up again allowed a resume from where it last took a snapshot... kind of virtual checkpoints
Just sharing for those who are willing to go the overlong mile to not loose much progress when a system bumbs or needs to boot mid job 3 and 12 [If you got multiple CEP2 running it could be even more interesting]. |
||
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1220 Status: Offline Project Badges: |
I can remember the first time I used a virtual machine to run tasks. It was before the windows application was launched. I can remember it took me ages to get the machine set up correctly. Rental machines are a wonderful invention. In my personal experience I find hibernating my computer if I need to turn it off much easier. Sounds like it could be interesting times ahead for the project. Thanks for bringing the idea to our attention Rob.
---------------------------------------- |
||
|
Paul Schlaffer
Senior Cruncher USA Joined: Jun 12, 2005 Post Count: 231 Status: Offline Project Badges: |
In the case of VMWare Workstation which I use, you can just suspend the virtual machine whenever a reboot is required. No snapshot is needed, although you can do so if desired. For myself I only reboot when installing patches, so I would likely run the WU as is and just plan my reboots. Sounds like a huge data set, bring it on!
----------------------------------------“Where an excess of power prevails, property of no sort is duly respected. No man is safe in his opinions, his person, his faculties, or his possessions.” – James Madison (1792) |
||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Ah yes, when there was the very long "Linux only" period, when it took serious effort to get it to run on Windows, [with a much more efficient result... Linux is no match to the Windows version, by exception]. A VM not requiring snapshotting is even better considering the size of the models middle of the job segments, but it's when there's uncontrolled shutdowns.
And not to be misunderstood, this particular discussion was not about CEP2, nothing of WCG in fact, but leeching operating [best] practices is never wrong. |
||
|
Rickjb
Veteran Cruncher Australia Joined: Sep 17, 2006 Post Count: 666 Status: Offline Project Badges: |
I tried running WCG in a VM under Windows, and found that there was considerable computational overhead in doing so. The WUs in the VM ran about 15% more slowly than directly under Windows.
So before you go ahead and use your VM, check the cost in speed and decide whether the benefit is worthwhile. My reason for trying the VM was to allocate a fixed number of CPU cores to CEP2; the VM device would have CEP2 as the only project in its device profile, while the device profile for the cores outside of the VM would run only other projects. Now that BOINC supports an app_config.xml file, such a setup is not needed. |
||
|
Randzo
Senior Cruncher Slovakia Joined: Jan 10, 2008 Post Count: 339 Status: Offline Project Badges: |
Hello Rickjb,
which virtualization platform have you used? What OS was used as guest and host system? It can depend on it a lot. Also you should give the VM enough RAM and disable swapping on it as far as it is extremely inefficient on VM. |
||
|
mmstick
Advanced Cruncher Joined: Aug 19, 2010 Post Count: 149 Status: Offline Project Badges: |
Wouldn't it be possible to apply a similar concept of simply dumping the cache in memory to disk instead of a virtual machine with overhead? That's really all virtual machines do when you save the state. However, I don't use winbloat since I am a Linux user.
----------------------------------------[Edit 1 times, last edit by mmstick at Jun 1, 2013 4:20:17 AM] |
||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I agree with Randzo. I have Ubuntu installed using VirtualBox on my windows machines. The Vina projects run about 20% faster for me, despite the overhead and I don't see much difference in CEP times. FAAH and HPF2 are about 10% slower. Just give it enough ram as accessing virtual memory from disk is a system killer.. VM or not:)
|
||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Another project I crunch for is already doing something similar:
----------------------------------------The LHC 2.0 project "Test 4 Theory" is testing BOINC's native VirtualBox wrapper. All of the project work units are run inside a Linux VM running on VirtualBox and BOINC coordinates the setup/teardown of the VM and balancing CPU utilization if there are other BOINC projects running. This makes things simpler on the researcher's end, since the science app has a uniform target platform and the suspend/resume behavior is handled by VirtualBox. The downside is that you lose some CPU performance due to virtualization overhead. -Steve edit: added important "on" [Edit 1 times, last edit by Former Member at Jun 5, 2013 1:16:18 AM] |
||
|
dango
Senior Cruncher Joined: Jul 27, 2009 Post Count: 307 Status: Offline Project Badges: |
Hello,
as mentioned in different topic (HPF2) I'm unable to receive work for these project (HPF2, CEP2) -> HPF is gone, so I need to solve problems with this project. I'm running Linux (debian) virtual machine as kvm qemu client. Output from cpuinfo: model name : QEMU Virtual CPU version 1.4.0 flags : fpu de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pse36 clflush mmx fxsr sse sse2 syscall nx lm rep_good pni vmx cx16 popcnt hypervisor lahf_lm in cc_config.xml I put this lines, but nothing happened <cc_config> <options> <alt_platform>i686-pc-linux-gnu</alt_platform> </options> Any other idea? Thanks |
||
|
|