Message boards :
Number crunching :
Points/Credit
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Send message Joined: 2 Oct 19 Posts: 21 Credit: 4,947,474 RAC: 1 |
You will have to take that up with sebastien as he came to the conclusion that if you run more clients (work units "intended by his standards") than actual CPU cores/threads then you were essentially "cheating". Basically, he didn't like people finding ways to run more NCI work units on rigs as he felt that was side stepping projects intentions and wishes even if the project encouraged the behavior. The real fear is that he doesn't want people running a ton of WUProp instances for "credit whoring" reasons when it isn't adding more "value" to his project stats reporting which are already skewed anyways. He sees no value to a thousand instances all reporting the exact same thing from the same host. However, people were trying to run WUProp in these instances because they wanted credit for the run time hours they actually DID run for the purpose of the badge upgrades. He adjusted the project so that it didn't work right when running multiple instances and tried to make a lame compromise. However, even the compromise was broken (not sure if it still is and don't care at this point). I see no difference on whether I run 2 instances or 1000 instances when discussing fairness of runtime hours. As far as preventing people from abusing the points...that is a different discussion with different solutions. He also made the change so that in order to get credit for the run time, you also had a CPU intensive work unit running on the same client. That is stupid because someone could still run a CPU intensive app and just tell it to run 10% of the time. Thus allowing other clients to still credit and not over commit the system. Around this time was also when he started to address people running projects that had NCI but didn't have the proper "sensors" to actually compute the work units like QCN and Radioactive. People were running the apps doing nothing in mass just to get run time hours for badges. People can still run a ton of Virtual Machines and waste a ton of resources in order to get WUProp credit bumps too. What it boils down to is that he just didn't like the amount of credit one could squeeze from an NCI project and was afraid they would do it to his. Simple as that. |
Send message Joined: 25 Aug 19 Posts: 409 Credit: 1,240,185 RAC: 0 |
The current approach in the NCI project and tasks is indeed problematic. The issue of current tasks and credit Windows/Linux is also problematic. The iThena PERF application is currently only available on Windows. I will add new applications only after the old tasks are finished (this is the plan). Currently there are ~107053 tasks for iThena PERF in line. Maybe this process will take 24 hours? I am aware of the problems. But you can't do everything in one day. Most of the problems (probably all?) can be solved but it takes time and a lot of work. |
Send message Joined: 2 Oct 19 Posts: 21 Credit: 4,947,474 RAC: 1 |
|
Send message Joined: 24 Sep 19 Posts: 108 Credit: 910,182 RAC: 182 |
Yes thank you Rysiu. We know that you are working very hard, thanks again. Conan |
Send message Joined: 6 Oct 19 Posts: 17 Credit: 265,766 RAC: 49 |
I now get PERF tasks on my linux hosts and they are rewarded a glorious 2 credits per task ! I now have a 18 RAC and 70 total credit ! Alleluia !! \o/ |
Send message Joined: 25 Jan 20 Posts: 24 Credit: 258,254 RAC: 46 |
|
Send message Joined: 6 Oct 19 Posts: 17 Credit: 265,766 RAC: 49 |
I really don't know, I didn't do anything special, in the project preference there is no application choice, they just started to arrive. I have 2 debian 10 and one debian 9 hosts and they all get 2 tasks (one of each application) running at the same time. |
Send message Joined: 25 Aug 19 Posts: 409 Credit: 1,240,185 RAC: 0 |
Tasks can be assigned at random. At this point the scheduler has not yet been modified. I'll probably modify it this weekend. |
Send message Joined: 24 Sep 19 Posts: 17 Credit: 1,588,258 RAC: 0 |
Tasks can be assigned at random. It has been my anecdotal experience that: 1) I did not start receiving the PERF tasks on Linux computers until *after* I filled out the information form for that Linux computer. 2) Whenever there's an outage where iThena is down, i.e., the server was restarted,, afterwards I do not get any PERF tasks on Linux computers until I resubmit the forms for those computers. It's entirely possible that this is a coincidence, or that my observations are flawed. Want to find one of the largest known primes? Try PrimeGrid. Or help cure disease at WCG. |
Send message Joined: 23 Sep 19 Posts: 9 Credit: 3,249,000 RAC: 40 |
I received all CNode tasks initially. Update via BOINC manager after the timeout got me PERF tasks. Did not resubmit the form. |
Send message Joined: 23 Sep 19 Posts: 25 Credit: 2,224,656 RAC: 218 |
I also see that after some time period of running both, some event stops dual running from continuing. I have not had to resubmit the form - i just made the client update, after the 5 minute wait after downloading a CNode tasks has expired, and then a PERF tasks downloads. For some period thereafter, both run and get new tasks happily. |
Send message Joined: 19 May 20 Posts: 2 Credit: 16,902,020 RAC: 0 |
New task seems to run much faster about ~140 sec each compare to 4000 sec plus previously. Is this a temporary thing? |
Send message Joined: 12 May 20 Posts: 29 Credit: 1,939,078 RAC: 120 |
Check out the front page news section on WUProp project to understand the implications of Goofyxgrid and it's effect on WUProp as seen by the Admin of that project. So you are saying I can set up a virtual machine for 1 less than the total cpu's I have in a machine and crunch Ithena in each one of them and be just fine? The same thing for WuProp? I'm assuming that does NOT include the HT or SMT cpu cores on Intel or AMD. mikey |
Send message Joined: 12 May 20 Posts: 29 Credit: 1,939,078 RAC: 120 |
The current approach in the NCI project and tasks is indeed problematic. Are you going to add a Mac app as well? |
Send message Joined: 25 Aug 19 Posts: 409 Credit: 1,240,185 RAC: 0 |
Are you going to add a Mac app as well? Mac applications are possible. However, we need to have a place/system to compile such applications and have time to test them. But a Mac application is possible. |
Send message Joined: 12 May 20 Posts: 29 Credit: 1,939,078 RAC: 120 |
Are you going to add a Mac app as well? I asked too soon...Apple is going to Arm chips of their own design and at least the mobile ones may not be able to crunch due to the power they would use. Seems the numbers of Mac users is dwindling and some projects are already discussing not continuing support for them. |
Send message Joined: 6 Oct 19 Posts: 17 Credit: 265,766 RAC: 49 |
I've been using my iMac from 2010 to 2020. This is one of the reason I bought one : I knew it would last, unlike any PC. And it did. So saying "stop everything Apple is changing to their own SoC" is quite... ridiculous. I'll take 2 years (they say) so the current product offer is "all converted to the new SoC", and many (many) years before Intel based Mac have disappeared from the surface of the earth. |
Send message Joined: 25 Aug 19 Posts: 409 Credit: 1,240,185 RAC: 0 |
There are a lot of possibilities here... It'll probably take some time to change the architecture... A lot can change... At this moment we're talking about a Mac for Intel processors? It probably won't be difficult to implement for a Mac, but I haven't tested it... |
Send message Joined: 12 May 20 Posts: 29 Credit: 1,939,078 RAC: 120 |
I've been using my iMac from 2010 to 2020. This is one of the reason I bought one : I knew it would last, unlike any PC. And it did. NO the idea was not to put energy into something that is going to stop at some point anyway, they do not have a Mac app here my thought was if it takes alot of effort the return may not be worth it. |
Send message Joined: 2 Oct 19 Posts: 21 Credit: 4,947,474 RAC: 1 |
What I have seen, is that VM's are less affected by the change than multi-clients are. However, you can run multi-clients inside a VM as well. His change pretty much target multi-clients. So, you can run as many VM's that you want but that would be a lot less efficient than running multi-clients. But it would allow for more WUProp instances. It was a dumb "solution". Someone could get away with loading a ton of VM's even if they over committed the CPU's if they really wanted to. I wouldn't recommend it, but it all comes down to the system and how you configure them I guess... |
© 2019-2024 iThena. All rights reserved. | Private Policy