Threadripper 3990x stingy scheduler

Message boards : Number crunching : Threadripper 3990x stingy scheduler
Message board moderation

To post messages, you must log in.

AuthorMessage
bossmaniac

Send message
Joined: 29 Nov 20
Posts: 1
Credit: 29,791,692
RAC: 45,207
Message 389 - Posted: 8 Jan 2021, 10:21:31 UTC

I did not get enough wu, only 20 for the queue on a threaripper 3990x. I run out of wu overnight serverial times. There is no resting timer running. I find my computer idling in the morning and i haven't yet configured a 0 prio backup project. The backup project wouldn't help to keep the Computer running with Sidock wu.
Please increase the limits of task in progress for this processor.
ID: 389 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MindCrime

Send message
Joined: 2 Jan 21
Posts: 1
Credit: 1,507,670
RAC: 0
Message 390 - Posted: 8 Jan 2021, 10:53:11 UTC - in response to Message 389.  

it's not just big and fast cpus. my little 3770k and 3570k run out often. it looks like the core count x2 is max at a time.

it would be nice if they could raise max in progress.
ID: 390 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
OCNfranz

Send message
Joined: 9 Jan 21
Posts: 2
Credit: 6,501,015
RAC: 7,439
Message 392 - Posted: 10 Jan 2021, 2:22:38 UTC - in response to Message 390.  

I am not seeing a 2 core limit on my Intel 4790K rig. Right now it is processing 6 tasks, 3 RxDock and 3 CurieMarieDock. My Ryzen 1700x rig has not received more than 1 task at a time though.
ID: 392 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jim1348

Send message
Joined: 30 Oct 20
Posts: 57
Credit: 9,112,528
RAC: 0
Message 393 - Posted: 10 Jan 2021, 12:11:32 UTC

I expect that they limit the number sent out to each machine in order to get them back quickly.
I can live with that, if it helps the science. Otherwise, I could use more too.
ID: 393 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
JagDoc

Send message
Joined: 24 Oct 20
Posts: 19
Credit: 10,234,681
RAC: 11,076
Message 401 - Posted: 12 Jan 2021, 8:40:44 UTC

The limit of 2 WU per core is not the problem.
But with the limit of 128 WU per host it is not possible to run hosts with 128 or more cores efficient at SiDock.
This value should increased.
ID: 401 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Buro87 [Lombardia]

Send message
Joined: 23 Nov 20
Posts: 28
Credit: 771,948
RAC: 0
Message 402 - Posted: 12 Jan 2021, 9:26:55 UTC - in response to Message 401.  
Last modified: 12 Jan 2021, 9:38:24 UTC

I think it would be better to rise it up to 512wu...
Some Threadripper 64c/128t are already present on this project
On others projects there are dual Epyc (128c/256t) monsters....
ID: 402 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
xii5ku

Send message
Joined: 3 Jan 21
Posts: 24
Credit: 30,967,539
RAC: 116
Message 419 - Posted: 17 Jan 2021, 10:33:21 UTC

For us donors who need to plan and maintain our donations, there are generally three methods to approach this:

1) Ask the project admins for a general increase of these host limits. Of course this is what this thread is for.

    Lately, this project became better in keeping small/slow computers fed with work overnight. Large/fast computers are still prone to run out of work, and as mentioned, more than 128 logical CPUs are not being fed even with 1-deep workqueue. 88- and 96-threaded computers have been a common sight in Distributed Computing for years now, 128-threaded computers are quite common now too, and 256-threaded computers are making more appearances too. Particularly, what the SiDock@home server is perceiving currently is listed here: cpu_list.php

    So the question is, to which degree are the admins ready to cater to the convenience of donors with larger/faster-than-average hosts?


2) As an individual donor of large/fast hosts, get in personal contact with an admin and ask for the server to have extra limits configured for selected host IDs.

    I don't know if this is acceptable here at SiDock. I never tried this approach at any project myself; I merely saw that some project admins do this.


3) Partition your single large/fast computer into several small/slow computers.

    This gives very good control into your own hands, and it works with almost all projects. Therefore, this is what I routinely do at projects with low limits of work in progress per host, if I am having trouble to maintain a reasonable work buffer at such projects. This method needs to be implemented responsibly though. I am not going into detail how to implement it.

ID: 419 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Havis

Send message
Joined: 2 Mar 21
Posts: 5
Credit: 25,783,600
RAC: 17,276
Message 799 - Posted: 19 Apr 2021, 10:01:02 UTC

looks like the problem is not software, but the disk space on the server, I think that one needs to be upgraded to higher capacity, so that all computers can be properly fed with workunits.
ID: 799 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Threadripper 3990x stingy scheduler

©2024 SiDock@home Team