The trend of hosts active in last 7 days graph provided by stef on the same web page as the other graphs posted in this thread shows a very encouraging reversal in the last few weeks.
Thanks. Hm, too bad, my own script overwrote the log instaed of appending to it, so the last day is missing again.
Quote:
Sorry I had no time yet to look for the problem, though I am pretty sure that a serverside provided "%/d"-value would solve the problem as well :-)
To avoid accumulating rounding errors I'd suggest to record the (integer) number of workunits that are done and derive the percentage done in the last day from that.
Is this showing the start of the University/School term and students using "free" (prepaid) electricity?
Or is the Northern hemisphere cooling down enough for people to fire up their crunching farms again?
I'd guess returning student effect for the several week old trend (61k to 67.5k).
The huge uptick in the last day, however, I suspect is a case of refugees from SETI, which is not distributing new work at the moment, as their science database is down. If the even larger uptick in percent per day is to be believed, this latest batch of users has faster than average hosts as well.
The 2-week graph was the correct one. The overall-graph was wrong, so I removed it.
It's strange though because both graphs are calculated the same way.
It's strange though because both graphs are calculated the same way.
I don't know if this has any bearing on reality, but I ran into several strange calculating and rounding situations on the Seti charts. A couple were plain old brain cramps on my part, others were related to the way RRD does things. An obvious example was the way RRD 'smoothed' values. If the actual readings from the source were "5,5,5,5,7,7,7,7", and these values were what were stored into the RRD, upon fetching that same data it would have transmogrified into "5,5,5,5,6,7,7,7". If the values were of larger magnitude, and changing more frequently, that 'intermediate transition' number resulted in a more obvious skewing of averages, max's and min's. When the raw data was used with pen and paper to calculate averages it came out considerably different, and more logically accurate, than what the 'built-in' averaging produced. One of the RRD developers had stated this behavior was by design due to the 'resolution' of data and the frequency of updates allowed, and to give displayed charts a more 'smooth' look rather than a 'square-wave' appearance. The more recent RRDTOOL versions have a "-no-slope" option that may play into all this. I have since converted over to using SQL and letting the scripts do the math, so I never pursued a solution or workaround from an RRD prespective.
The trend of hosts active in last 7 days graph provided by stef in its "last two weeks" version appears to show a temporary uptick which may be related to the 9-hour SETI outage earlier today. It appears that at least a few heavy-duty SETI crunchers run relatively short queues, and divert very substantial resources to Einstein rather quickly after an outage.
It is also quite visible in the percent per day graph, which has been oscillating in the .400 to .430 range, and at this moment of writing has spiked up to .475. For tidiness I'll not patch that graph in, but in case you've recently joined this topic, stef's longer-term graphs may be found at
Stef: RE: Here is the
)
Stef:
Would you mind to update it for me as of, say, today?
I just started logging the values myself and want to close the gap.
Best,
Bernd
BM
Ok:http://homepage.hispeed.ch
)
Ok:
http://homepage.hispeed.ch/einstein/einstein3.xml.bz2
Sorry I had no time yet to look for the problem, though I am pretty sure that a serverside provided "%/d"-value would solve the problem as well :-)
Please send me a mail if you find something.
5htt [at] paranoya [dot] ch
RE: The trend of hosts
)
Is this showing the start of the University/School term and students using "free" (prepaid) electricity?
Or is the Northern hemisphere cooling down enough for people to fire up their crunching farms again?
What's the best guess?
Happy crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
RE: http://homepage.hispeed
)
Thanks. Hm, too bad, my own script overwrote the log instaed of appending to it, so the last day is missing again.
To avoid accumulating rounding errors I'd suggest to record the (integer) number of workunits that are done and derive the percentage done in the last day from that.
BM
BM
RE: Is this showing the
)
I'd guess returning student effect for the several week old trend (61k to 67.5k).
The huge uptick in the last day, however, I suspect is a case of refugees from SETI, which is not distributing new work at the moment, as their science database is down. If the even larger uptick in percent per day is to be believed, this latest batch of users has faster than average hosts as well.
Here is a recent
)
Here is a recent log:
http://homepage.hispeed.ch/einstein/einstein4.xml.bz2
The 2-week graph was the correct one. The overall-graph was wrong, so I removed it.
It's strange though because both graphs are calculated the same way.
RE: It's strange though
)
I don't know if this has any bearing on reality, but I ran into several strange calculating and rounding situations on the Seti charts. A couple were plain old brain cramps on my part, others were related to the way RRD does things. An obvious example was the way RRD 'smoothed' values. If the actual readings from the source were "5,5,5,5,7,7,7,7", and these values were what were stored into the RRD, upon fetching that same data it would have transmogrified into "5,5,5,5,6,7,7,7". If the values were of larger magnitude, and changing more frequently, that 'intermediate transition' number resulted in a more obvious skewing of averages, max's and min's. When the raw data was used with pen and paper to calculate averages it came out considerably different, and more logically accurate, than what the 'built-in' averaging produced. One of the RRD developers had stated this behavior was by design due to the 'resolution' of data and the frequency of updates allowed, and to give displayed charts a more 'smooth' look rather than a 'square-wave' appearance. The more recent RRDTOOL versions have a "-no-slope" option that may play into all this. I have since converted over to using SQL and letting the scripts do the math, so I never pursued a solution or workaround from an RRD prespective.
The trend of hosts active in
)
The trend of hosts active in last 7 days graph provided by stef in its "last two weeks" version appears to show a temporary uptick which may be related to the 9-hour SETI outage earlier today. It appears that at least a few heavy-duty SETI crunchers run relatively short queues, and divert very substantial resources to Einstein rather quickly after an outage.
It is also quite visible in the percent per day graph, which has been oscillating in the .400 to .430 range, and at this moment of writing has spiked up to .475. For tidiness I'll not patch that graph in, but in case you've recently joined this topic, stef's longer-term graphs may be found at
http://homepage.hispeed.ch/einstein/
and the last two week variants at:
http://homepage.hispeed.ch/einstein/index_p.html
An alternate interpretation is that the uptick started before the SETI outage, and may relate to some other source.