40 WU/day limit - what's the deal?
Message boards : Problems and Bug Reports : 40 WU/day limit - what's the deal?
I a have nice little computer farm; one of my computers is an overclocked AM3 Sempron with a 550Ti GPU. For starters, BOINC client made a huge mistake while benchmarking the system; it continuously sends very little work and I have to manually update the connection with the server every day lest I run out of work - the computer is much, much faster than BOINC benchmarking would suggest.
|ID: 115534 · Reply Quote|
You have more than 50 Einstein units in progress. What do you want more? I am running 7 BOINC projects, including two Virtual Machines, with a 0.25 day cache and I never get a new unit until I finish and report the preceding. I am running SuSE Linux 11.1 on an Opteron 1210 at !.8 GHz. no GPU.
|ID: 115535 · Reply Quote|
... BOINC client made a huge mistake while benchmarking the system ...
BOINC benchmarks your CPU only. If you pair up a decent GPU with a slow, single core CPU, you may have less than optimal conditions for efficient crunching, particularly with this project. The CPU is needed for a significant proportion of the time for the BRP CUDA tasks so this will be noticed when you also crunch CPU only tasks on that single core. That GPU would benefit from a fast multiple core CPU.
... it continuously sends very little work and I have to manually update the connection with the server every day lest I run out of work ...
The E@H project has a download limit of 40 tasks per CPU core per day. Your GPU can do a large proportion of that limit by itself so it's not surprising that you will tend to run out of work. You would solve your problem if you had a dual core CPU because you would then be able to download 80 tasks per day.
... the computer is much, much faster than BOINC benchmarking would suggest.
Actually it's probably not. When running CPU only tasks, your host is only doing a few per day.
Today, however, I increased the "additional" WU cache to 10 days in BOINC client, hoping that the server would have some mercy, sending me more work units and saving me from having to manually update the client ...
The project will only send you 40 tasks per day no matter how hard you plead :-). Hopefully, you have (close to) zero for the "connect to internet every X days" (CI) setting. Also there is no need to set the "extra days" as high as 10. It would probably hinder you if you have a significant non-zero CI as well.
... BOINC, for some reason, uploads the completed WUs but doesn't do the update/acknowledge part to the server automatically - or it waits for 12+ hours to establish a connection with the server, during which I run out of work.
BOINC is designed to be economical with server contacts. Results upload immediately but reporting is delayed so that multiple results can be reported in a single contact. That single contact is usually when extra work needs to be requested. If your BOINC client is not requesting work even when your cache is almost fully exhausted (and you haven't reached your 40 tasks limit) you should look carefully for what is causing BOINC to think you don't need more work. Something seems a bit strange there.
This afternoon I was greeted with the message from the server; "exceeded daily quota of 40 WUs". Well I'm pretty darn sure I did less than 40 WUs on that computer on that day, so that left me puzzled.
There's no puzzle. How many tasks you have actually done is not important. That message is all about how many tasks have actually been sent to you already on the current day.
I'm just about to run out of GPU work units, so this is a bad scenario.
I believe Sempron 145s are actually dual core Athlons with one core disabled. It is possible to unlock the extra core (with the right motherboard). Google it. Maybe the extra core is faulty but in many cases it's not - luck of the draw, I guess. With two cores you'd get 80 tasks/day.
Also, gamma-ray WUs take too long to crunch and return way too little credit for the processor time invested. Can something be done about that?
Browse your tasks list for that host and look for completed FGRP tasks. Notice the big difference between CPU time and elapsed time. This is because you have only one core which is being significantly used by all your CUDA tasks. The FRGP tasks are being prevented from getting all the CPU they need for efficient crunching. You would see much better FGRP performance if you had something like an Athlon II X4 or better (if you wish to stay AMD) paired with your GPU.
|ID: 115544 · Reply Quote|
Thanks for clearing that up =)
|ID: 115545 · Reply Quote|
|ID: 115546 · Reply Quote|
Message boards :
Problems and Bug Reports :
40 WU/day limit - what's the deal?