Also the data were taken for all host with credits ( ~ 2 million), but only < 100,000 host are really active right now.
The real number of Opterons could be way lower than the reported "host with credit" number.
Example: I have a triple-boot-system (linux,win xp 32, win vista64). Just last month I changed the hard disk and made a clean install of all 3 OSs. In the first year I also had vista32 and not vista64.
On all of this OSs I installed BOINC and let it finish at least a few tasks.
So my Core2Duo will appear in this statistic probably as 10 host, but is really only one computer.
If they did this scenario with a cluster, this could explain the huge number of Opterons.
Indeed interesting, both for the curves and for the number of samples for each CPU type.
The Intel i5 looks to be anomalous... Some quirk there?
The most error prone part of this code is taking the vendor and model strings in the download and translating them into the categories. I've done enough spot checking to show first numbers but I will look much more closely at the I5 when I get back home (It's now 0830 local).
Quote:
The great shame is that the cobblestones benchmark is so far divorced from any useful approximation to reality...
I'm wondering if there is general public (me) access to the work unit records. I think a more useful comparison would be between individual processors on the same work unit.
Now that's still pretty far divorced from anything else I use my CPUs for but it is better than the benchmarks.
Regarding i5: they're available as dual cores with HT and a few of them are quad cores without HT. So far the latter is a minority (i5 750 and 760), so the un-overclocked ones of these may represent a "high-performance spike" in your data.
Regarding database access: you may want to take a look at AndyK's homepage. He's got a script running which shows the pending credits, so that's at least some form of external database access.
Regarding database access: you may want to take a look at AndyK's homepage. He's got a script running which shows the pending credits, so that's at least some form of external database access.
MrS
I didn't see any details of Andy's script but I can guess how it works.
I've been reluctant to "web crawl" for the data I want, pretty much out of courtesy but if the PTB don't mind it's doable. I could even throttle it to a reasonable bandwidth, not that my home cable ISP could put much of a load on UWM's Internet connection.
I've been reluctant to "web crawl" for the data I want, pretty much out of courtesy but if the PTB don't mind it's doable. I could even throttle it to a reasonable bandwidth, not that my home cable ISP could put much of a load on UWM's Internet connection.
I don't recall a position statement from Einstein officials, but the SETI folk are definitely opposed to what they call "scraping". Their attitude seems to be that automated stuff should use their published XML, and as to what is available on user pages but not in the XML, that is for human, not automated access.
To be more specific, they asked a few high-profile users who were making regular use of AndyK scripts to keep track of their own work to stop, and they did stop. (quite likely others still use them).
Indeed interesting, both for the curves and for the number of samples for each CPU type.
The Intel i5 looks to be anomalous... Some quirk there?
The most error prone part of this code is taking the vendor and model strings in the download and translating them into the categories. I've done enough spot checking to show first numbers but I will look much more closely at the I5 when I get back home (It's now 0830 local).
I looked at which cpu's get identified as I5 and I don't believe any were misidentified. At least my program agrees with my brain. I couldn't think of an efficient way to check if any were misidentified as something else.
Quote:
Regarding i5: they're available as dual cores with HT and a few of them are quad cores without HT. So far the latter is a minority (i5 750 and 760), so the un-overclocked ones of these may represent a "high-performance spike" in your data.
This makes a lot of sense. I'm working on separating the data by family, model, stepping. I've generated the histograms, graphs should be out in a day or two. Summary of I5 looks like:
I don't recall a position statement from Einstein officials, but the SETI folk are definitely opposed to what they call "scraping". Their attitude seems to be that automated stuff should use their published XML, and as to what is available on user pages but not in the XML, that is for human, not automated access.
To be more specific, they asked a few high-profile users who were making regular use of AndyK scripts to keep track of their own work to stop, and they did stop. (quite likely others still use them).
That's how I treat my sites. The effect of crawling or scraping is similar to a DoS attack but with better motives.
I can present some data in a
)
I can present some data in a more general format.
The first try divides the hosts by vendor and model name. I will get histograms by family, model and stepping next.
Let me know if you find it interesting.
Joe
RE: I can present some data
)
Indeed interesting, both for the curves and for the number of samples for each CPU type.
The Intel i5 looks to be anomalous... Some quirk there?
The AMD plots are pretty much as would be expected. One interesting point is the number of Opterons! Is some big datacentre kindly helping out?
The great shame is that the cobblestones benchmark is so far divorced from any useful approximation to reality...
Keep searchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
RE: The AMD plots are
)
The number 1 user the Nemo-cluster was build with operons but now the most credits come from the new Xeons
http://einstein.phys.uwm.edu/top_users.php
http://einsteinathome.org/account/144041/computers
http://www.lsc-group.phys.uwm.edu/beowulf/nemo/index.html
Also the data were taken for all host with credits ( ~ 2 million), but only < 100,000 host are really active right now.
The real number of Opterons could be way lower than the reported "host with credit" number.
Example: I have a triple-boot-system (linux,win xp 32, win vista64). Just last month I changed the hard disk and made a clean install of all 3 OSs. In the first year I also had vista32 and not vista64.
On all of this OSs I installed BOINC and let it finish at least a few tasks.
So my Core2Duo will appear in this statistic probably as 10 host, but is really only one computer.
If they did this scenario with a cluster, this could explain the huge number of Opterons.
RE: Indeed interesting,
)
The most error prone part of this code is taking the vendor and model strings in the download and translating them into the categories. I've done enough spot checking to show first numbers but I will look much more closely at the I5 when I get back home (It's now 0830 local).
I'm wondering if there is general public (me) access to the work unit records. I think a more useful comparison would be between individual processors on the same work unit.
Now that's still pretty far divorced from anything else I use my CPUs for but it is better than the benchmarks.
Joe
Regarding i5: they're
)
Regarding i5: they're available as dual cores with HT and a few of them are quad cores without HT. So far the latter is a minority (i5 750 and 760), so the un-overclocked ones of these may represent a "high-performance spike" in your data.
Regarding database access: you may want to take a look at AndyK's homepage. He's got a script running which shows the pending credits, so that's at least some form of external database access.
MrS
Scanning for our furry friends since Jan 2002
RE: Regarding database
)
I didn't see any details of Andy's script but I can guess how it works.
I've been reluctant to "web crawl" for the data I want, pretty much out of courtesy but if the PTB don't mind it's doable. I could even throttle it to a reasonable bandwidth, not that my home cable ISP could put much of a load on UWM's Internet connection.
Joe
RE: I've been reluctant to
)
I don't recall a position statement from Einstein officials, but the SETI folk are definitely opposed to what they call "scraping". Their attitude seems to be that automated stuff should use their published XML, and as to what is available on user pages but not in the XML, that is for human, not automated access.
To be more specific, they asked a few high-profile users who were making regular use of AndyK scripts to keep track of their own work to stop, and they did stop. (quite likely others still use them).
RE: RE: RE: Indeed
)
I looked at which cpu's get identified as I5 and I don't believe any were misidentified. At least my program agrees with my brain. I couldn't think of an efficient way to check if any were misidentified as something else.
This makes a lot of sense. I'm working on separating the data by family, model, stepping. I've generated the histograms, graphs should be out in a day or two. Summary of I5 looks like:
(Sorry I don't see a way to get fixed pitch fonts on this BB, code usually does it).
Joe
RE: I don't recall a
)
That's how I treat my sites. The effect of crawling or scraping is similar to a DoS attack but with better motives.
Joe
RE: [pre] +--------+-------
)
Try (with [] of course)
Grüße vom Sänger
