LHC@home

BOINC tries to send the fastest app for your computer, so if say avx was fastest then ti would try to do this. It also throws in a few other other to test if your computer got faster on the other types ever now and again, if it was found to be faster then it would switch over.

There is quite a few caveats here so it does seem quite random in the end.

BOINC tries to send the fastest app for your computer, so if say avx was fastest then ti would try to do this. It also throws in a few other other to test if your computer got faster on the other types ever now and again, if it was found to be faster then it would switch over.

There is quite a few caveats here so it does seem quite random in the end.

Are these created for sse2, avx, etc in advance? So it can be possible all the avx have run out from the queue and you have to be given the slower one?

On the subject of queues, it appears a recent upgrade to LHC has made it hand out the right subprojects. It used to be that if I selected Atlas, and others if no work is available, that even though there are thousands of Atlas on server status, I would always get Theory. I'd have to specifically exclude Theory and untick allow other apps also. But now I get Atlas even by asking for "anything".

---

On the subject of queues, it appears a recent upgrade to LHC has made it hand out the right subprojects. It used to be that if I selected Atlas, and others if no work is available, that even though there are thousands of Atlas on server status, I would always get Theory. I'd have to specifically exclude Theory and untick allow other apps also. But now I get Atlas even by asking for "anything".

Actually it seems that now it is only sending Atlas and no Theory or sixtrack. At least that is what I am seeing here. I would like to run mostly Atlas but my smaller host don't have the memory to do so. Only two at the time is maximum and if I have Atlas enabled the Host only downloads them and nothing else. This leaves CPU cores idling. I have to disable Atlas and abort a few of the previously downloaded ones to force downloads for Theory and sixtrack.

I currently have 5 Atlas tasks (2 running and 3 waiting). I disabled Atlas on project preferences and managed to download 8 sixtrack tasks but still Boinc don't want run sixtrack unless I suspend the Atlas tasks in queue. So it would like to run three Atlas tasks simultaneously but don't have the memory to do so. But it will not start sixtrack either unless I suspend the Atlas tasks that are waiting to run. Once it has started sixtrack tasks I can resume the Atlas tasks in my queue and it will run the sixtrack tasks to end until it should start new tasks again but won't do so. So rinse and repeat.

---

Actually it seems that now it is only sending Atlas and no Theory or sixtrack. At least that is what I am seeing here. I would like to run mostly Atlas but my smaller host don't have the memory to do so. Only two at the time is maximum and if I have Atlas enabled the Host only downloads them and nothing else. This leaves CPU cores idling. I have to disable Atlas and abort a few of the previously downloaded ones to force downloads for Theory and sixtrack.

I currently have 5 Atlas tasks (2 running and 3 waiting). I disabled Atlas on project preferences and managed to download 8 sixtrack tasks but still Boinc don't want run sixtrack unless I suspend the Atlas tasks in queue. So it would like to run three Atlas tasks simultaneously but don't have the memory to do so. But it will not start sixtrack either unless I suspend the Atlas tasks that are waiting to run. Once it has started sixtrack tasks I can resume the Atlas tasks in my queue and it will run the sixtrack tasks to end until it should start new tasks again but won't do so. So rinse and repeat.

An admin in here once told me something about "it's whatever's in the memory buffer". I more recently heard that the Boinc server software is rubbish, and the queue to send out to us volunteers has to be in RAM, and not on a disk, so if the tasks are big, LHC could have a problem if they don't have huge amounts of RAM on the server. I guess they have to guess how many of each will be requested and put those in RAM. If they judge slightly wrong, you have to make do with something else. Not sure why it won't accept a massive swapfile on an NVME....

---

I don't know what the problem is, but I feel that LHC needs a bit too much micro managing to get it running properly with several subprojects at the same time.

---

I am gettting very short sixtrack tasks, both sse and avx, on a Windows 10 PC. Atlas tasks complete regularly on two cores, CMS tasks fail every time.
Tullio

---

I think as the others said I think they are generated and there is separate queues for each app.

You can see the speeds for all the apps on my computer here https://lhcathome.cern.ch/lhcathome/host_app_versions.php?hostid=9930009

I gave up on this, I just let it send me whatever it likes, I wanted to have even balance between the sub-projects but it's not really possiable

I have the best luck using two machines, with CMS and native ATLAS on one, and Theory and SixTrack on the other.

I often can not get Theory by itself. It seems that it needs SixTrack too.
And CMS is often out, so you need ATLAS to fill in for it. (It takes a lot of memory though).

I gave up on this, I just let it send me whatever it likes, I wanted to have even balance between the sub-projects but it's not really possiable

Same here. If I get fed up of seeing one all the time, I just ban that one for a while so I get the others.

---

I am gettting very short sixtrack tasks, both sse and avx, on a Windows 10 PC. Atlas tasks complete regularly on two cores, CMS tasks fail every time.
Tullio

I only have problems on a very old (DDR2 era) PC. The rest can do any of the LHC subprojects. And I've never seen your two core problem. Virtualbox both 5 and 6 both work fine here (had to downgrade it for another project....)

---

I have no problem on Atlas tasks on the 3 cores Windows 10 PC. Those on an old Linux laptop take a very long time and do not produce a HTTS file.
Tullio

---