I've been considering for some time creating a Rosetta benchmark that would have a set of known tasks that one could run over and over again. Each time using exactly the same starting random seed etc. That way the results should be absolutely consistent from one run to the next, then changes and tweaks can be further studied for their effects to specific benchmarks.

Basically (once I get it done), all you would do is attach to a new "project" that would send you the tasks that make up the benchmark, so maybe 5 tasks. It would then watch for your results and record your reported runtimes on the tasks. And I might create several different benchmarks with a unique subdirectory name on the project, so you could connect to mydomain.com/benchmark1 or mydomain.com/benchmark2, and compare and discuss with others that run the same benchmarks.

You would get zero credit for doing the tasks. They would basically be tasks that the project sent out in the past and has already received resutls from. But this would let us review the actual credit granted for the models produced as well.

Once I work through all the kinks, I could do the same for other projects as well. I basically plan to just harvest an actual scheduler reply and send it to each client the attempts to connect to my "project".

Not positive if I'll be able to rework things around to change the platform from that of the originally harvested client. If that can be done without corrupting signatures, then you could actually run the tasks on Linux, reboot the machine to Windows and run the exact tasks again there on the same machine.

Before I undertake the endeavor, I was wondering if people think that such benchmarks would be useful, and if they'd need to be able to create their own benchmark sets of tasks etc. for it to be useful, or if some "stock" set of benchmarks would serve the purpose. What do you think?

---

Add this signature to your EMail:
Running Microsoft's "System Idle Process" will never help cure cancer, AIDS nor Alzheimer's. But running Rosetta@home just might!
https://boinc.bakerlab.org/rosetta/

To me, that sounds like an excellent idea. The ability to compare a machines performance so that it produces the best result is never a bad thing.

You'd be able to compare different OS's, CPU's, motherboard's, RAM.

As for the amount of benchmarks, as long as there was a wide enough range it'll be fine. I've no idea how the client really works but maybe if there were different run times for the data so you could see which has the potential for the most work done.

Back to my original point, I've changed my machine so it only gets 1 day of work and see if that makes any difference. It'll allow me to compare like for like in case it makes a difference.

---

Well I've updated the machine with Vista 64 and the newest BOINC client yet sadly no change.

link

I'll order a HT 3.0 motherboard on monday and hopefully that will make a difference.

---

Related question:

if you reboot a machine does the task time go back to the last checkpoint, or does the time remain and just the work reset to the last checkpoint?

if it's the latter, the credit will be lower per cpu second...

---

Related question:

if you reboot a machine does the task time go back to the last checkpoint, or does the time remain and just the work reset to the last checkpoint?

if it's the latter, the credit will be lower per cpu second...

I presume the first but it won't matter either way as it's on 24/7 without rebooting. I've got a dual core that's been on for 54 days now without a reset :)

---

Related question:

if you reboot a machine does the task time go back to the last checkpoint, or does the time remain and just the work reset to the last checkpoint?

if it's the latter, the credit will be lower per cpu second...

I presume the first but it won't matter either way as it's on 24/7 without rebooting. I've got a dual core that's been on for 54 days now without a reset :)

ah

probably not the issue then ;)

---

Related question:

if you reboot a machine does the task time go back to the last checkpoint, or does the time remain and just the work reset to the last checkpoint?

if it's the latter, the credit will be lower per cpu second...

I presume the first but it won't matter either way as it's on 24/7 without rebooting. I've got a dual core that's been on for 54 days now without a reset :)

Is the RAC going down on that machine? Windows has 'issues' with not releasing stuff as it does its thing, meaning less and less is available for other programs, ie Boinc in this case, making them slow down. I tend to reboot mine about once a month, manually running an anti-spyware program at the same time.

---

Is the RAC going down on that machine? Windows has 'issues' with not releasing stuff as it does its thing, meaning less and less is available for other programs, ie Boinc in this case, making them slow down. I tend to reboot mine about once a month, manually running an anti-spyware program at the same time.

Nah the RAC has never gone above 500 but it hovers around 480 ish. It's got masses of spare memory too as it's a dual core running on 2GB of RAM.

---

Rather interestingly, my RAC is picking up now even though I'm using the old client. Looks like whatever I did, it's working :)

---

Rather interestingly, my RAC is picking up now even though I'm using the old client. Looks like whatever I did, it's working :)

Or you didnt have to do anything? :) My Core i7 had problems overtaking another Core i7 in the top hosts list a few weeks ago, it stayed behind for days, RAC was then around 3900, before overtaking and continued to climb to above 4100 before leveling off, the other i7 stayed at around 3900 and is still there. I believe its like mod.sense said here (and dcdc), "credit is not an absolute thing. It is based on how difficult all the other machines out there are finding it to produce models for a given task" - and thats always changing and the RAC change with it. R@H credit system is imho inaccurate but its about as good as it gets without quorums,

---