hi tvdsluis,

It may have coincided with a 'heavier' batch, but yes the increase on my quad was about 10-15 minutes average up to 4.6-4.7 hours, but opposed on my C2D it came down by almost half an hour. The overall project WU average is coming down suggesting that those with truly extended hours have cut it by a very significant amount.

Within a series I've observed variations from 3.8 to 4.9 hours, so for now am considering it within norm, but interested to here if more see this on a continuous basis. The programmers may have had to compromise somewhere to work around the page faulting which for some was in the billions and doubling run time.

Here's some stat data given the daily average hours per job.

7,413227 feb 18
7,376983 feb 19
7,313571 feb 20
7,126219 feb 21
7,185980 feb 22
7,140714 feb 23
7,200180 feb 24

What's important is that overall averages are lower and consistent across all platforms and that seems to have been achieved.

There is a supplemental graphs post that tracks daily averages. Anyone keeping an eye out can see there how each project is moving.

ttyl

Hello tvdsluis,
We had a problem overloading the servers when Uplinger produced some very short FAAH work units. This was compensated for initially by reducing the percentage of FAAH units sent out and increasing HPF2 and HCC. Then I saw some of my work units getting about an hour longer. Not a big difference, but I figure that they are allocating more work per unit to reduce the load on the server. It seems to be a small increase. Once FAAH settles down, they might shorten it a bit again.

Lawrence

Here a chart showing the marked improvement of average run time, particular where the version 5.15 comes out of the system and is substituted with results processed using the 5.20 science release. Typically since the slower machines are last to report, there is often a brief spike before dailies resume their regular averages. Similar at the start of a series one sees the population to be initially only composed of the fast devices delivering first after which the slower begin to add to the daily averages.

To come back to the duration, the amount of work in these jobs is I think constant, but the Chrystal image could from one to the next set be holding more information to analyze, all chrystal 'quality' not being equal.



(From the ego stroking department ;>)

I got several HCC-units today with an estimated time of 19 hours, what's up with that? Normally these units are crunched in 5 hours or so.

Ok they are in the queu to be run, let's see how long it will take to crunch them.