Hello arrizza,
I do not understand your question. Your figures say that you produce 3296.06 points per day on a Virtual Machine but 5495.20 points per day on a Physical Machine. That sounds right to me -- Physical is faster than Virtual.

What am I missing?

Lawrence

First things first, there is VB enabling development ongoing. See wiki:http://boinc.berkeley.edu/trac/wiki/VirtualBox. A fair chunk to read and digest. [If you see the name 'Kevin' than many will know who that is (tech of WCG, currently at another annual BOINC workshop)].

A CPU has so many hertz. If x cycles are used to manage a VB instance that is to run a BOINC client on the same computer as the BOINC client that is already running in the ''real'' OS environment, then there will be less cycles for sure going towards work unit computing. In run-time the tasks will use more time, but in CPU time, one measure by which BOINC assigns credit, less time will be recorded.

If the virtual environment is used to run it onto a second computer, than the gain is not far from 100%, all depend on what that second computer does. That would be my aim.

After reading the wiki, some jaws will be a little distance lower. :D

--//--

P.S. Thought there'd be a news release of some kind (Lawrence made a brief note on the forums IIRC), but WCG servers on the work-handout side are now running in an internal cloud environment.

As SekeRob has already explained a given physical machine can give a finite number of processing cycles per calendar day. If you have BOINC installed in this physical machine there cannot be many unused cycles.
A virtual machine will certainly use more than these few unused cycles, and anything in excess will be less cycles available for BOINC.
In total both BOINCs will have less cycles to use than a single one in the physical machine.

There are two things (at least) which confuse you when you do your measurements:
1. you probably use runtime as reported by BOINC, and we all know that this is wrong, same as when BOINC runs two tasks in an hyperthreaded processor. BOINC (and the OS) report two days of CPU time per calendar day for that processor, which is simply wrong (by much ).
2. in WCG it is not possible to use "one result" as a meaningful unit of work done: one result can need from a few minutes to many hours, even in the same machine. WCG points or BOINC credits cannot be used either because the yield per unit of CPU time is too disparate from one project to another.

If you want to get a better picture of the productivity of your configurations you should
1. select a single WCG project for all your BOINCs, ideally Computing for Clean Water which has the most consistent run times
2. count the number of tasks completed** per calendar day, ideally during several days.

My opinion: if your objective is to improve your ranking in the runtime classification, then install as many BOINCs as you can inside the same physical machine,
but if your objective is to produce the most useful work per day, then keep one single BOINC at the physical level.

** Completed, not validated: although CCW tasks are usually validated immediately this is only about 90 % true because about 10 % are distributed to two clients and can be "pending validation" for some time.

I believed it was clear that my words about piling up virtual BOINCs in one single physical machine were ironical, especially after having tried to help arrizza to evaluate the real efficiency of his/her choices.

Have you forgotten that I have never been enthusiastic about WCG's celebrating its efficiency in centuries of runtime, although I recognise that this is the only thing which means something to an outsider, unfortunately.

Hi Guys,
I share totally Jean's meaning regarding the "value" of Time even if it is probably the most understandable figure for outsiders as well as for non technical people.
I think that every cruncher should be honest in some ways and that artificially "stretching" respectively "increasing" participation does not bring any advantage to the community and to the scientists.
Indeed every overhead - and virtualization generates overhead - is on the cost of the application (in this specific case on Boinc costs, in particular for "Boinc only hosts"). For this reason, it is simple to understand that running Boinc within a virtualized environment (even if multiple instances are up at the same time) will never bring more performance that running Boinc on the physical host.
I would only see a justification for virtualizing Boinc in computer farms and/or on blades in order to not jeopardize in any way the stability of the physical hosts but however taking advantage of the host idle time.
Cheers,
Yves