Daily quota
10 May 2018 17:03:24 UTC
Topic 214911
(moderation:
I can't get enough tasks!!
The issue I'm having is with host 5116827. It has a quad core i5-3570K CPU and a Vega 64 GPU capable of completing a (valid) GRBP task every 3 minutes or so. - My cache settings are to store enough work for 1 day + 0.1 days (so not greedy).
I'm hitting a daily download quota for new work on this host of 384 tasks (which is presumably calculated on the CPU core count. My other Vega toting machines are i7's and don't have an issue). However, the host in question by my calculation is capable of completing somewhere in the order of 480 tasks per day and is daily sitting idle for an extended period waiting for the project backoff to count down before it gets new work...
Perhaps the project admins could re evaluate the limits for reliable hosts, please :-)
Gav.
Your card is a lot fater than
)
Your card is a lot fater than my 660Ti SC but I still have mine set this way so I always have work here on my older quad-core.
https://einsteinathome.org/host/12001170
And I run tasks X2 with this pc and use the other 2 cores to run Cern VB task(s)
I admit that I load up with a few hundred and then set the Boinc Manager to not get new tasks and just reload about a week later.
Thanks for responding Magic.
)
Thanks for responding Magic. Unfortunately increasing the number of days to cache work will not help with this host as the server is limiting me to a total download of 384 tasks per day, which is not enough to fill my 1 day plus the additional 0.1 days cache settings.
As I said originally, I believe the download limit for any given host varies according to the amount of reported CPU's that are available. So unless there is a way to convince Boinc that I have more CPU's than just the 4 available, the daily download limits will have to be adjusted server side...
I had crunched through all my available tasks for this host by mid morning today and no matter how many times I did a project update to try and get new work the server said no! and I would have to wait another 14 hours before being able to get new work.
My rough and ready work around was to temporarily swap this machines Boinc directory into one of my i7 machines to download a days worth of tasks whilst masquerading as an 8 core i7 machine ;-) before swapping the directory back to this machine to continue crunching.
The i7 machines have a daily download limit of 512 tasks (for those that are interested) and I guess the described procedure is something that I will have to perform every couple of days to maintain work flow for this machine :-)
Gav.
Well that is strange
)
Well that is strange Gavin
On my old quad-core where I just run the tasks X2 (and use the other 2 cores for Cern 2-core tasks)
And I have got over 500 Einstein tasks (when I accidentally left it set to allow new tasks)
When we first started running the BRP binary search #1 on GPUs v1.20 version I got 534 before I reset to *no new tasks*
I don't want to get too many so I leave it set *no new tasks* but I just tested mine and now I have 439 running and probably would have got more if I didn't stop it.
https://einsteinathome.org/host/12001170/tasks/1/0
But then yours is running the GPUs v1.18 version so maybe........
And I don't know if you only having 4GB ram makes any difference (mine has 12GB)
And mine is still using Win7 and you have Win 10
Sure is strange that you can do that much work and not get more automatically as you send in the finished tasks.
Good luck Gavin
Gavin_14 wrote:... I believe
)
Hi Gavin,
I did a search for "daily AND quota" and got lots of hits over lots of years. I started looking through a few but quickly became frustrated simply because of the amount of time it was taking for a clicked link to actually display. It's probably a combination of an overloaded server and a slow internet connection but I gave it away thinking there has to be an easier way to work out how the daily quota is calculated. And it seems there is :-).
You mentioned that your quad core/vega host with the 384 daily quota is returning a task every 3 mins. The task list shows ~360 secs so you are running 2x. I was pretty sure that the quota depended on both the CPU and GPU numbers but I wasn't sure if GPU task concurrency had any influence. I upped the work cache on one of my dual core hosts with an RX 570 running 3x. I wanted to use something that had a different concurrency as well as different CPU core count.
Turns out it's daily quota is 288. The figures that give a sensible solution are that the per CPU core quota is 48 tasks and the per GPU extra is a further 192. GPU task concurrency doesn't seem to be involved. The other thing to note is that if you were crunching CPU tasks as well (I realise you are not), you would have to subtract that number from the quota to get the allowed GPU tasks per day.
I know none of that helps you but if you could convince BOINC that you had 8 cores instead of 4, your quota would increase by a further 192. A fair while ago (probably before there were the restrictions on the top hosts list) I remember seeing hosts listed that seemed to have ridiculous numbers of cores. Perhaps there is a way to manipulate the core count. Maybe somebody hand-crafted a version of BOINC to do that :-). However, I think your best option would be to send a PM to Bernd to try to get his attention. It's a bit silly for you to be restricted in this way.
Cheers,
Gary.
MAGIC Quantum Mechanic
)
That's a direct consequence of having such a large cache setting (10 + 10) that you posted earlier. The daily quota would limit you to 384 new tasks per day. Because you are not completing anything like 384 tasks per day, the tasks you actually have on board will just keep increasing day by day, by the difference between 384 and what you actually do complete. This will continue until you remember to set NNT or until BOINC decides that tasks won't make the deadline and so it goes into panic mode. It's not strange at all. You are just giving BOINC an impossible task to manage without ongoing user micro-management.
I don't understand why you would want to force yourself to have to remember when to turn the tap on and when to turn it off again, repeatedly like this. Why don't you work out what cache size you want and just set it? Perhaps it is something forced on you by what LHC needs? I don't know anything about LHC these days but surely it doesn't require you to maintain a 10+10 cache setting??
Also, I can assure you that the behaviour you see and the behaviour Gavin sees are nothing to do with the app version, the brand of GPU, the RAM size (I believe his is actually 8GB) or the Windows version. Unfortunately, the daily quota system isn't smart enough to be able to relax the limits just because all those tasks being returned are 'good' tasks and not 'bad' tasks.
Cheers,
Gary.
Gary I am not saying I have
)
Gary I am not saying I have any problems here at all.......I am just telling Gavin what I do with the only GPU card I run these days on a similar quad-core.
I do it this way because I want it to run the Einstein tasks without ever worrying about it running out of work and yes I do spend most of my time with the Cern projects these days (I do the Alpha and Beta testing) and the regular public versions.
For a few years I did run all 10 of my computers here since they all still have the Geforce cards but the cost of power made me just retire all of them except the 660Ti SC and it makes no sense for me to replace all my 650Ti,550Ti and 560Ti cards to the current new versions to do this here.
(I am sure you have spent plenty of money doing this but I am at the point in life where I see no reason to spend thousands of dollars just to get *credits*)
But I will keep my 660Ti running until it croaks and for the first time in many years decided to test run some CPU tasks here on this laptop 8-core that has been running since 2011 since its GeForce card just died and I just did a clean install of the OS (it does run the CPU temp a bit high so I added an exterior fan)
I get more credits here with just the one card but I run 50 cores for Cern since they are all CPU and I test the original multi-core and other Alpha tests and LHC-dev.
Gary Roberts wrote: A fair
)
Eureka!!
Cheers Gary, now you've jogged my memory I too recall those hosts with 200+ CPU's and the best bit is that there is no voodoo magic involved to increase the reported core count, Boinc actually gives you the tools to do it in cc_config with the use of the <ncpus> flag :-) Host 5116827 now has 8 cores reported and I can now download enough work to get me through the day :-)
Thanks again Gary for the nudge in the right direction.
Gav.
Gavin_14 wrote:Thanks again
)
You're most welcome! I'm very glad you've got it sorted :-). It must have been a bit painful having to swap the whole BOINC tree between hosts in order to top up all the time. I bet you're glad that's over!
I've never needed to use cc_config.xml so am not really familiar with all the options. I've read through the list of them on occasion but my brain doesn't retain things like it used to, particularly if I don't have an immediate need for that thing. Now that you mention it, I do recall seeing that option before and wondering why anyone would need to fudge the number of cores. I guess I'll remember this one now that I see a good use for it :-).
Cheers,
Gary.