That might just be due to the windows x86 credit being targeted at the nonSSE app, with the SSE version being roughly twice as fast. While the creditwhore's have all gone elsewhere in search of manhood substitutes the mass of ordinary boxes with DCFs dropping from 1 to .5 when they went from the stock s4 to the sse s5 apps triggered the performance bump.
I rather think that slow hosts (PII, K6 and older) went to projects with less demanding WUs but the remaining hosts do more Flopses with the default optimisations now.
edit : The summer definitely is a "problem", CPU heat is not so welcome in those days - well, and the summer heat is not so welcome to the CPUs either
The positive trend did not continue. Unfortunately it's back to going backwards.
6+ days later the number of days remaining is only down by 3.5 days and total days needed has gone way up again. I wish that wouldn't happen all the time.
Total needed 406.5 days with 379.2 days remaining as of a few minutes ago.
Quote:
Posted 9 Jul 2006 3:23:23 UTC
Moving forward. The trend is positive again. That's better.
Total needed 403.9 days with 382.7 days remaining.
Quote:
On 06 Jul 2006 3:51:30 UTC Server Status is:
Total needed 406.0 days with 387.7 days remaining. Roughly one week has gone by and days remaining has increased by 10.4 days.
Quote:
On 29 Jun 2006 0:16:33 UTC Server Status was:
Total needed 388.4 days with 377.3 days remaining.
I've been hoping this would get better as the fraction of effort devoted to S4 work dies down. One might think that benefit is nearly all already reflected, as the number of S4 units outstanding with non canonical result is down to 292 as I type.
However it appears that the scheduler is making increasingly strenuous efforts to bring in the last few of these stray sheep.
I have two S4 returned results for which credit is pending.
For this one, the initial results were sent out on June 16. The potential third result errored on July 4. Since then the scheduler has sent out six more, of which one promptly errored but the other five remain "in progress". five out there
This second one is nearly the same story, save that the first error out happened much earlier another five out there
It might be that some slow-responding hosts will still be crunching on this flurry of late S4 completion attempts for as much as a couple of weeks, though it is hard to think they represent a very big fraction of total capacity.
I had one machine (my slowest) which happened to be sharing a datafile with a bunch of folks who dropped out. It was still crunching S4 exclusively until early July 12.
It may be a month or two before both the real situation and the estimation method settle down. Of course if something new happens in the optimization arena before then, settle time may be yet longer.
That might just be due to the
)
That might just be due to the windows x86 credit being targeted at the nonSSE app, with the SSE version being roughly twice as fast. While the creditwhore's have all gone elsewhere in search of manhood substitutes the mass of ordinary boxes with DCFs dropping from 1 to .5 when they went from the stock s4 to the sse s5 apps triggered the performance bump.
Summertime has arrived, many
)
Summertime has arrived, many turn off there extra computers and than restart them in the fall when the heat from the CPU is needed. (only kidding)
EDIT
That would only be a munite part of it, I do see the visitors at my web site drop every July & Aug. than go up higher than before in the fall.
Try the Pizza@Home project, good crunching.
It seems that the drop of
)
It seems that the drop of 1,000,000 credits a couple days ago was due to Bruce.
Bruce
Einstein Credit
I rather think that slow
)
I rather think that slow hosts (PII, K6 and older) went to projects with less demanding WUs but the remaining hosts do more Flopses with the default optimisations now.
edit : The summer definitely is a "problem", CPU heat is not so welcome in those days - well, and the summer heat is not so welcome to the CPUs either
Moving forward. The trend is
)
Moving forward. The trend is positive again. That's better.
Total needed 403.9 days with 382.7 days remaining.
The positive trend did not
)
The positive trend did not continue. Unfortunately it's back to going backwards.
6+ days later the number of days remaining is only down by 3.5 days and total days needed has gone way up again. I wish that wouldn't happen all the time.
Total needed 406.5 days with 379.2 days remaining as of a few minutes ago.
I've been hoping this would
)
I've been hoping this would get better as the fraction of effort devoted to S4 work dies down. One might think that benefit is nearly all already reflected, as the number of S4 units outstanding with non canonical result is down to 292 as I type.
However it appears that the scheduler is making increasingly strenuous efforts to bring in the last few of these stray sheep.
I have two S4 returned results for which credit is pending.
For this one, the initial results were sent out on June 16. The potential third result errored on July 4. Since then the scheduler has sent out six more, of which one promptly errored but the other five remain "in progress".
five out there
This second one is nearly the same story, save that the first error out happened much earlier
another five out there
It might be that some slow-responding hosts will still be crunching on this flurry of late S4 completion attempts for as much as a couple of weeks, though it is hard to think they represent a very big fraction of total capacity.
I had one machine (my slowest) which happened to be sharing a datafile with a bunch of folks who dropped out. It was still crunching S4 exclusively until early July 12.
It may be a month or two before both the real situation and the estimation method settle down. Of course if something new happens in the optimization arena before then, settle time may be yet longer.
RE: It seems that the drop
)
Bruce uses only 640 Opteron 175 systems to crunch WUs... :)
( ~ 2500 WU / day )
ture, that still means he was
)
ture, that still means he was responsible ~40-50% of the hit.
actaully, your estimate of
)
actaully, your estimate of his WUs/day is a bit off. Acording to boincstat's he's averaging ~600k credit per day, which is ~3400 175cr WUs.