bwoochowski
Member
- Mar 25, 2004
- 38
- 14
- 81
Grats everyone on one billion credits in PrimeGrid
voodoo5_6k
Senior member
- Jan 14, 2021
- 395
- 443
- 116
Hmm, "Einstein@Home | No work is available for Gamma-ray pulsar binary search #1 on GPUs" does make it a slow process queueing Einstein-work...
By my recollection the other type of GPU-wu's are FUBAR and wherefore not an option.
But, on CPU, is where any kind of pros/cons between the 3 different types of Einstein-work, or doesn't it really matter?
By my recollection the other type of GPU-wu's are FUBAR and wherefore not an option.
But, on CPU, is where any kind of pros/cons between the 3 different types of Einstein-work, or doesn't it really matter?
E@H is basically a GPU project. CPU credit does not compare to the GRPB #1 on GPUs app. Its not really worth running anything else.
- Oct 9, 1999
- 4,328
- 2,112
- 146
Upload/report all Universe@home bunkers. If you are only running Uni, run it wide open right now.
Upload/Report all NumberFields@home bunkers. If you ran NF on the previous throws, continue to run it open now.
If you are new here, attach to both projects above and run whatever tasks you can get.
Likewise, if you are running Uni and can't get tasks then run NF.
If you are running NF and can't get task, then run Uni, but try to get tasks on NF and switch back ASAP.
All open running GPUs in PrimeGrid.
Bunker Einstein@Home on GPUs, if you can't get E@H tasks then bunker SRBase, if either are out of work then run PG.
Upload/Report all NumberFields@home bunkers. If you ran NF on the previous throws, continue to run it open now.
If you are new here, attach to both projects above and run whatever tasks you can get.
Likewise, if you are running Uni and can't get tasks then run NF.
If you are running NF and can't get task, then run Uni, but try to get tasks on NF and switch back ASAP.
All open running GPUs in PrimeGrid.
Bunker Einstein@Home on GPUs, if you can't get E@H tasks then bunker SRBase, if either are out of work then run PG.
I can't get my Windows node to add the Universe@Home project and I have been trying for days! Is there anything I am missing? I tried the command line and the GUI and nothing... I always get communication error. My Linux machine is crunching fine Universe@Home
Also, how to bunker? You're talking about delaying result submission until they start counting points, correct? What are the configs to delay submission?
Also, how to bunker? You're talking about delaying result submission until they start counting points, correct? What are the configs to delay submission?
voodoo5_6k
Senior member
- Jan 14, 2021
- 395
- 443
- 116
Yes, sir, I'm on it
Aborting PG, now working on SRBase bunker, dropping once the sprint starts! CPUs stay on Uni. Have a bunch of unreported tasks, uploading them right now.
voodoo5_6k
Senior member
- Jan 14, 2021
- 395
- 443
- 116
You might want to read about multiple clients to make bunkering easier.
How To - Multiple boinc clients on the same computer | AnandTech Forums: Technology, Hardware, Software, and Deals
Then, it's basically a matter of getting tasks, and don't allowing uploads (therefore, either set the BOINC networking config to 0.01 KB/s max upload and allow network usage only between 00:00 and 00:01, or block the respective server's IP via your firewall or translate them to localhost via /etc/hosts). And then release the tasks at the right time by undo ing whatever you did to block uploading. That's basic bunkering.
Not sure. I had been adding a few Linux hosts during the past few days and it always succeeded right away. Did maybe the initial download of the Windows version of the BHspin application fail? I don't have a Windows computer at Universe myself.
BTW, U@h's Linux application performs roughly twice as fast (or even faster?) compared to the Windows application on the same hardware. That's certainly a case of missing optimizations in the Windows application version.
What @voodoo5_6k said, and:
The basic way of bunkering is to fetch some amount of work, then prevent network communications, get the work done, and re-enable network communication when desired. (There are several ways to prevent network communication: Most commonly it's done by setting "Activity -> Suspend network activity" in boincmgr, or the equivalent in boinctasks/ boinctui/ boinccmd. Or you set a bogus HTTP proxy address in boinc's "other options". Or you assign the IP address 127.0.0.1 to the the project server's hostname in /etc/hosts or in a local DNS server of yours. Or you block the project server in a firewall. Or you pull the Ethernet plug.¹)
A more advanced way of bunkering is to fetch work, then get the work done and the results files transferred but suppress the reporting of the results all the while, and finally trigger the reporting when desired. The background: When a task was completed, the application saves its results into files (six files per result in case of the BHspin application), then the boinc client uploads these files to the project server, and after that's done, the client issues a "scheduler request" to the project server in which it reports all currently completely uploaded results. Only after that report can the project server act on the results, i.e. validate them. (In case of Universe@Home, the validation happens by comparing results of the same workunit which came in from two different hosts. That is, your result can only be validated after some other computer, commonly called "wingman", reported its corresponding result too.)
Since about the 2nd day of the Pentathlon, the Universe@Home server's internet link is at or very near its limit for file uploads to the server. Therefore, it's difficult to upload a large amount of result files at once, like it would need to happen with the "basic way" of bunkering. Hence, at least everybody who performs some substantial bunkering should rather implement the "more advanced way" of bunkering.
Though if you suppress communications like I described at the "basic way", you would prevent both the uploading and reporting. Letting the uploads through but retain the reports can be done in a few, more involved, ways:
- Block access to the scheduler server of the project, e.g. by an entry in /etc/hosts, but leave access to the upload server(s) of the project functional. This is only possible at a small number of boinc projects (those which have a separate host name or even host address for the scheduler), but Universe@Home is one of these few projects where this works.
- A method which works with most (perhaps all) projects is to perform several settings in the client, then enable network activity and let it upload, then disable network again and revert the settings. What settings and other preparations need to be done somewhat depends on the particular project. Among else, to do this right, one should suspend the project in the client for the duration of the uploads. Hence, a downside of this method is that the client can't compute while it uploads files.
- For completeness, another method would be to use a boinc client which was compiled from customized source code. I never tried that myself. There are only few users in the entire boinc community who use one or another source code modification of the client, and I don't know if any of them made modifications specifically for the purpose of enabling uploads while retaining reports.
________
¹) As @voodoo5_6k noted, there is furthermore the method of pseudo-blocking the transfers by configuring a very low upload rate. I never used that method myself, because a) it is obviously much less reliable than the other ways, b) with a bit of planning, this method can be avoided and one or more of the other methods used.
Last edited:
PS:
As today is an "obstacle day" = "bonus day" at the Pentathlon's Obstacle Run, there is probably a whole lot users who employed "basic bunkering" and are trying to upload a lot at once today; same as on the previous obstacle day last Friday. Therefore I expect it to be rather futile throughout much of today — if not well into tomorrow — to try such things like adding a new host.
pschoefer, one of SETI.Germany's Pentathlon organizers, posted several comments on this year's project choices: SG forum post, in German. (I didn't check yet whether popular machine translators produce a sensible translation of this post.) Earlier he already noted in the shoutbox that MilkyWay@Home was contacted during the Pentathlon preparations, but did not respond.
waffleironhead
Diamond Member
- Aug 10, 2005
- 6,919
- 429
- 136
Its good they didnt choose MIlkyway, they have been having difficulty for the last month or more just keeping the project affloat and normal users fed. The admin has been dealing with HD issues and server problems.
Anyone just tuning in the NF javelin throw is almost over so make sure you switch all CPUs to Universe@Home. Run them dark if you can (bunker) or run them open if you're not able to bunker.
If I recall correctly, the 1st Obstacle caused uploads to be virtually impossible during the entire May 6 and maybe 2/3rds or more of May 7.
Project May 5May 6May 7May 8May 9May 10May 11May 12 May 13 May 14 May 15 May 16 May 17 May 18 Final Obstacle Run 200%
150%
Given the long period from the between 1st and 2nd Obstacle, it is likely that uploads will again be near impossible during most of today, if not even way into tomorrow. (This clog is of course caused by those users who deferred uploads until the obstacle. As noted before, the more successful approach here is to upload as soon as practical, and only defer the result reporting.)
After the current clog is cleared, there aren't very many days left until the third, final clog. The 3rd clog should be somewhat shorter than the current one, but if it is scheduled late, it could last until the very end of this contest.
Unfortunately, computing + uploading very early, while deferring the result reporting, involves substantially more management than by far most DC contributors are used to, or have spare time for. Personally, I am in a comparably good position to go through the motions of this sort of operation: The bulk of my production comes from dual-socket computers which tend to complete the work which one client instance can buffer very quickly. That is, a quick staggered cadence of computation + upload comes natural with bigger computers like these. Therefore I am already used to buffering with, say, a two-digit count of client instances per computer. I have some scripts for that, and I made arrangements to have opportunities throughout the day to look after my computers for the entire duration of the Pentathlon.
On the other hand: After having done multi-day/ week-long bunkering with several fast computers during past contests, I actually planned to perform only minimal bunkering in 2022's Pentathlon — just the amount of bunkering which lets me avoid the usual rush hours when a sub-contest starts, allows me to be asleep when a contest starts, and allows for a decent validation ratio in quorum-2 projects. Alas, these personal plans became obsolete when the new Obstacle Run rules were announced. :-(
- Oct 9, 1999
- 4,328
- 2,112
- 146
Congrats team we took home Gold on Cross-country.
Please report to Discord for our next moves.
Please report to Discord for our next moves.
voodoo5_6k
Senior member
- Jan 14, 2021
- 395
- 443
- 116
I for one don't use discord et al. Don't know, but maybe we could also have like group PM with the brief strategy outline, so everybody stays informed? Just an idea.
I try to post strategy stuff on here also. If you don't have access PM me and I'll get you set up.
Www.teamanandtech.org
voodoo5_6k
Senior member
- Jan 14, 2021
- 395
- 443
- 116
Thanks
I have access already. It just looked so quiet there during the first pentathlon days that I didn't look into there recently. I'll drop by Thanks
I'll do better at cross posting whats going on in Discord on the forums also.
The engraver went busy right away and took less than 15 minutes after the contest to manufacture the medals: :-)
https://www.seti-germany.de/forum/content/1514-BOINC-Pentathlon-2022-Cross-Country-Medal-Ceremony
TRENDING THREADS
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 10K
-
Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)- Started by Vattila
- Replies: 13K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 7K
-
Facebook
Twitter