I have 5 different hosts, 4 different OS's, CUDA GPU's
and 5 types of CPU's including PPC and Cell, and configuration
options for memory and disk that might make a significant difference to running Rosetta (and other
BOINC science apps). Chiefly, I'm trying to make
a decision of allocating apps to the hosts where they
run most efficiently. I think that credits per
hour is the relevant measure, but I don't see that
maintained anywhere. It would have to be per host,
not per project, and every BOINC project would need
to maintain the stats. Is that information available?

---

I have 5 different hosts, 4 different OS's, CUDA GPU's
and 5 types of CPU's including PPC and Cell, and configuration
options for memory and disk that might make a significant difference to running Rosetta (and other
BOINC science apps). Chiefly, I'm trying to make
a decision of allocating apps to the hosts where they
run most efficiently. I think that credits per
hour is the relevant measure, but I don't see that
maintained anywhere. It would have to be per host,
not per project, and every BOINC project would need
to maintain the stats. Is that information available?

I have never seen it and it may not be possible, each host can be doing multiple things other than running Boinc and that can affect the numbers. Also the amount of memory, the speed of the memory, the overclocking of the system, speed of the hard drives, etc, etc, etc, in addition to the things you already brought up, will all affect those numbers. However each project sets their credits based on a machine running at x speed, some actually have that machine, most just guess. If you could get those stats then you could do a spreadsheet that would be applicable across projects, but not really be real life numbers.

You also have to take into account how project a relates to project b and how they interact. I am running 2 projects that when run together seem to slow down the cpu one but the gpu one runs just fine. Either one run alone runs just fine but when the gpu one is running, the cpu project definitely slows down!

Hi Orchid

So you want a list of credits per hour for each host on Rosetta? You probably already know, and as mikey says, you can't compare the credits/hour from one project to another, but within a project your system should work well.

I don't believe the info you're after is presented on the rosetta website, and BoincStats should give you the relevant info but I've just checked and all mine are reporting 0.000000 against 'Average credit per CPU second' there.

You might have more luck on some of the other stats websites though - maybe gridrepublic? Otherwise you could copy a couple of pages from the stats here and divide the granted credit by the time taken...

---

I think what they are saying is they want to attach each machine to the project that will yield the most credit per hour for that machine. In other words, "which projects should these machines be attached to in order to maximize total credit granted per hour?" So that accomodates some of the variations that were mentioned.

I believe the credit per CPU second is in the exported host stats for each project. It is a huge file because it has all of the hosts attached to the project. But the stats sites already download it try to make the data presentable. Can anyone point us to one that presents the credit per CPU second?

---

Rosetta Moderator: Mod.Sense