// DBOINCP-300: added node comment count condition in order to get Preview working ?>
ohiomike
Joined: 4 Nov 06
Posts: 80
Credit: 6,453,639
RAC: 0
11 Jun 2007 14:06:46 UTC
Topic 192835
(moderation:
)
The last 5 WU's I have completed have not even been sent to anyone else yet. What's up with this new development?
PS- Not that I NEED the 2178.35 credits, but...
Datapacks are only sent initially to a limited number of hosts for processing. Depending on the speed of your wingmen and/or how many of them have dropped out of participation for the set you're currently running there can be a significant delay at times.
Typically during the 'mainstream' phase of a run, the project will decide eventually that progress on that set is too slow and start sending the set out to other hosts to get things moving again. On the earlier S5 runs you really didn't notice the delay all that much, but so far my experience on S5R2 has been that all those subtle effects have become very apparent now. ;-)
As I understand it, the problem is that the workunits you have received belong to a datapack that is kind of "rare" at the moment, so to speak (not many other hosts that have asked for more work seem to have them, yet???).
BOINC could force others to download this datapack (causing additional download traffic for those hosts) or wait for someone with this datapack to finally ask for new work (causing delay for your validation). So it's not trivial to find the right compromise between all the parameters you want to optimize (validation delay, download traffic, ...)
There seems to be a 7 days threshold for unsent results, at least I've never seen the "oldest unsent result" on the server stats page exceed 7 days significantly.
One other thing about S5R2 is previously the datapacks were a 'monolithic' ~30 MB monster which was DL'ed rather infrequently (except at the end of the run).
Now they consist of roughly 3.5 MB files and there are about 10 of them, so the 'working set' is still in the 30-35 MB ballpark. Also as you work through the templates, they get replaced with other ones as needed.
My guess this was done to reduce the BW demands on both ends and expedite the 'cleanup' at the end of the run, rather than an attempt to tailor the work to specific platforms.
It's just that clients get (rather huge) datapacks that are good for several workunits.
Yeah, they appear to be "good for" quite some time. I had a lengthy gap (9+ days) between the last 4 results I submitted and the 4 that I have to work on right now. I was expecting a server delete request and a new download, but the new results still pulled from the same data pack.
You will be pleased to learn, I'm sure, that I just got one validated on the 7th June, that was issued to me on 3rd May that I returned on the 5th May.
But still waiting for unit issued 16th May, returned 18th.
This unit was issued to my computer the 3rd of may. I returned it the 6th of may. Since then, 1 client error for that unit, 2 results not returned on time, and the current result in progress may time out the 14th of june. I'm not too optimistic for getting credits on that result anytime soon. ;)
WU's not sent
)
Datapacks are only sent initially to a limited number of hosts for processing. Depending on the speed of your wingmen and/or how many of them have dropped out of participation for the set you're currently running there can be a significant delay at times.
Typically during the 'mainstream' phase of a run, the project will decide eventually that progress on that set is too slow and start sending the set out to other hosts to get things moving again. On the earlier S5 runs you really didn't notice the delay all that much, but so far my experience on S5R2 has been that all those subtle effects have become very apparent now. ;-)
Alinator
Wingmen? That is the problem.
)
Wingmen? That is the problem. These five WU's were sent to me over 2 days ago, and to no one else as of today.
http://einsteinathome.org/workunit/33947858
for example.
RE: Wingmen? That is the
)
As I understand it, the problem is that the workunits you have received belong to a datapack that is kind of "rare" at the moment, so to speak (not many other hosts that have asked for more work seem to have them, yet???).
BOINC could force others to download this datapack (causing additional download traffic for those hosts) or wait for someone with this datapack to finally ask for new work (causing delay for your validation). So it's not trivial to find the right compromise between all the parameters you want to optimize (validation delay, download traffic, ...)
There seems to be a 7 days threshold for unsent results, at least I've never seen the "oldest unsent result" on the server stats page exceed 7 days significantly.
Am I getting this right??
CU
BRM
It looks to me that Mike just
)
It looks to me that Mike just moved up in template frequency, so it could be there just aren't that many other wingmen his current datapack set.
In this case it's just a matter of time until other hosts finish up theirs' and get assigned to the set he's working on.
As far as the oldest unsent goes, IIRC, that's usually about the oldest they get under 'normal' circumstances from what I've seen.
Alinator
Oh- this is part of the
)
Oh- this is part of the scheme to match WU's to PC abilities I take it?
RE: Oh- this is part of the
)
No, there is no such matching done in S5R2 AFAIK.
It's just that clients get (rather huge) datapacks that are good for several workunits.
I'm not sure how many different datapacks there are (maybe ca 10k??, 1 pack for every 0.05 Hz??) so they won't be distributed all at once , I guess.
CU
BRM
One other thing about S5R2 is
)
One other thing about S5R2 is previously the datapacks were a 'monolithic' ~30 MB monster which was DL'ed rather infrequently (except at the end of the run).
Now they consist of roughly 3.5 MB files and there are about 10 of them, so the 'working set' is still in the 30-35 MB ballpark. Also as you work through the templates, they get replaced with other ones as needed.
My guess this was done to reduce the BW demands on both ends and expedite the 'cleanup' at the end of the run, rather than an attempt to tailor the work to specific platforms.
Alinator
RE: It's just that clients
)
Yeah, they appear to be "good for" quite some time. I had a lengthy gap (9+ days) between the last 4 results I submitted and the 4 that I have to work on right now. I was expecting a server delete request and a new download, but the new results still pulled from the same data pack.
You will be pleased to learn,
)
You will be pleased to learn, I'm sure, that I just got one validated on the 7th June, that was issued to me on 3rd May that I returned on the 5th May.
But still waiting for unit issued 16th May, returned 18th.
Andy
LOL! This unit was issued
)
LOL!
This unit was issued to my computer the 3rd of may. I returned it the 6th of may. Since then, 1 client error for that unit, 2 results not returned on time, and the current result in progress may time out the 14th of june. I'm not too optimistic for getting credits on that result anytime soon. ;)
http://einsteinathome.org/workunit/33481656
And I'm quite certain there are users who have been waiting longer than I. :D