Whatever the performance difference is due to, it's not WU length. I looked at his active machines. The 3 celeron 2gigs are all geting ~40-50credit WUs.
This's completely different than my 2k/xp experience which showed little to no difference. MY guess is that the win boxes have other junk running in the background hogging CPU time. I'd suggest checking taskmangler for junk taking a share of CPU time, or checking each boxes private summary page to see what percent of the time boinc's getting the processor.
Just been reading this thread - haven't been here for a while. I have tweaked XP as far as I could, turning services off, registry tweaks etc, and have been running a dual-boot with assorted flavours of Linux. I don't have exact figures but if memory serves me correctly (??) using the old WUs from Einstein gave me the following results:
on 2.6 Intel with 768 meg RAM
Windows XP (service pack 2 added after install): around 8 hours 10 minutes
Kubuntu (Ubuntu with KDE GUI): around 7 hours 30 minutes
DSLinux: around 7 hours 15 minutes
SUSE 9.1 Linux: around 7 hours 45 minutes
(NB: I have reason to believe from reviews that I have read that SUSE 10.1 when it is released, will be somewhat faster than 9.1, tho' by how much I don't know)
I am running Gentoo Linux at the moment dual-booting Windows XP, but cannot comment on th WU times now as the changeover to assorted sizes happened before I installed it.
Just been reading this thread - haven't been here for a while. I have tweaked XP as far as I could, turning services off, registry tweaks etc, and have been running a dual-boot with assorted flavours of Linux. I don't have exact figures but if memory serves me correctly (??) using the old WUs from Einstein gave me the following results:
on 2.6 Intel with 768 meg RAM
Windows XP (service pack 2 added after install): around 8 hours 10 minutes
Kubuntu (Ubuntu with KDE GUI): around 7 hours 30 minutes
DSLinux: around 7 hours 15 minutes
SUSE 9.1 Linux: around 7 hours 45 minutes
(NB: I have reason to believe from reviews that I have read that SUSE 10.1 when it is released, will be somewhat faster than 9.1, tho' by how much I don't know)
I am running Gentoo Linux at the moment dual-booting Windows XP, but cannot comment on th WU times now as the changeover to assorted sizes happened before I installed it.
Hope this is of some use
Gray
Just to add more detail: in Windows I was running Zonealarm firewall and NOD32 antivirus in the days when those figure were applicable. An antivirus obviously affects computational power, and Zonealarm, while VERY neccessary, unless behind a router, is also a substantial drag on performance.
KDE in linux is defintely getting faster, but a window manager as supplied by DSlinux - Fluxbox, I think - is far lighter in its cpu requirements. If I was to build a box just for BOINC in the future, I would probably use Gentoo Linux and do a GUI-free install. Why Gentoo ? Hehe, I'm a control freak - pure and simple.
If there are processes stealing cpu time in XP, It is not of my doing. Each install is fresh out of the box with OS defaults and NO applications installed.
XP has a built in firewall that I will check the next time I am in the lab.
After this amount of time I am of the opinion that we could rule out different WU sizes as they should have averaged out by now.
The CPU efficiencies are vastly different machines and I was wondering just what that means and how to tweak it up a notch on the MS OS's. this statistic seems o be pretty relevant to the overall performance.
3 more machines are available for experimentation, please give suggestions as to which Linux versions would be best to compare. I am thinking of taking down Moe and Curly as the performance is just pathetic on a dedicated box and I think that the point has been proven and the test should be considered over. Once I tell them all to cease taking further WU's, I will compare the work done and make a conclusion when all of the WU's are verified. You are all welcome to look at the machines and compare.
Non GUI Linux...I would like to, but I may not have the technical skill as I am a noob. Perhaps that can be Experiment #3.
If there are processes stealing cpu time in XP, It is not of my doing. Each install is fresh out of the box with OS defaults and NO applications installed.
XP has a built in firewall that I will check the next time I am in the lab.
After this amount of time I am of the opinion that we could rule out different WU sizes as they should have averaged out by now.
The CPU efficiencies are vastly different machines and I was wondering just what that means and how to tweak it up a notch on the MS OS's. this statistic seems o be pretty relevant to the overall performance.
3 more machines are available for experimentation, please give suggestions as to which Linux versions would be best to compare. I am thinking of taking down Moe and Curly as the performance is just pathetic on a dedicated box and I think that the point has been proven and the test should be considered over. Once I tell them all to cease taking further WU's, I will compare the work done and make a conclusion when all of the WU's are verified. You are all welcome to look at the machines and compare.
Non GUI Linux...I would like to, but I may not have the technical skill as I am a noob. Perhaps that can be Experiment #3.
Hello there
There are a number of sites on the web that give info and stuff like services and other tweaks for XP - you might want to have a look here: http://www.theeldergeek.com/ and do a search for services. You might want to turn off all the GUI effects as well if you are using the box for anything barring crunching. Should you need an antivirus, grab either AVG, which has a free version or NOD32 which looks to have a good virii locating ability, but costs money - both are comparatively low on system resources.
If you have a broadband connection such as adsl with a router (wireless is rather more tricky I believe), Gentoo provides a decent set of documentation which you can access as you install. You'd need to get a minimal cd from one of their mirrors. This is quite small and just figures your dhcp connection plus hardware detection. It also gives you a text browser called Links with the address to their online docs supplied on the screen once you are in their command-line interface. I'm possibly a slightly advanced Win XP user, but I'm typing this from the Gentoo side of a Dual Boot system so it ain't THAT hard. Be warned tho' READ THE MANUAL !! Don't make any assumptions ! The forums are good too there, tho' they expect you to at least have tried to find the info before asking... and you can access them through links. The only issue timewise is the kernel compilation (use Genkernel if a noob - safer I feel) as that takes around 45 minutes on my 2.6 box. The other compilations are small, unless you want to go for a GUI, where the big files start... (Oh and if you use "grub" as a bootloader between XP and Gentoo, make sure you setup gub in (hd0) NOT (hd0,0). The latter is the usual windows partition and I bolixed my system that way before - late night installs are not really a good idea....
If you don't want to go through the Gentoo install, have a look at Damn Small Linux - only around 50 megs, has a GUI with some basic apps. It is designed as a live CD, but can be installed to a linux partition - maybe grab a Gentoo minimal CD which will allow you to make partitions fdisk /dev/hda for usual single drive box, and mke2fs -j /dev/hda(partition number) (without brackets) to give you an ext3 file system (one linux version of for instance fat32 for windows). If you have a router it picks it up and I had BOINC running easily in the dsl normal users home folder.
Just a slight update - I installed Gentoo Linux around 2 weeks back and now have dropped a BOINC binary into the system. I have just finished getting the Albert application and associated files and have begun to crunch - will be interesting to see if there is any significant speedup - especially as I have not really made any attempt to keep the system small and lightweight - I'm looking for a windows replacement so it has to do everything that windows used to do.
Obviously there will not be an exact measurement as there was with Einstein, more's the pity, but still worth the try anyhow - quite aside from me helping science in the process
Someone else stated that the assumption, Linux faster than Windows, only holds
on Intel CPU's. Couldn't find an example though. Maybe someone else
finds more examples.
Found a WU. But I don't like the outcome :(
For the record. Both processors are AMD XP 2600+. Mine (#464612) running Linux did the WU in 5,739.24s. The other (#48772) running XP 4,529.41s.
I think this difference comes from the counterclock signals. Probably the other cpu is overclocked. (XP2600+ factory speeds: 1917MHz,2080MHz,2083MHz,2133MHz)
The Results are consistently approx. 14000s (Win32) vs. 16000s (Linux)
(but still much better than the ~50-60% Linux disadvantage we saw last year with the older Clients)
It's indeed difficult to compare AMD CPU's with different Cores, as Einstein seems to care only for raw clockspeed (appears to fit easily into 256k L2 Cache or less).
Performance is also relatively lower on low clocked Athlon64 CPU's (e.g. 1800MHZ So939 Athlon64 3000+, which is easily slower than faster clocked AthlonXP's on the Einstein client but excels fine according its Performance Rating in other Projects)
Whatever the performance
)
Whatever the performance difference is due to, it's not WU length. I looked at his active machines. The 3 celeron 2gigs are all geting ~40-50credit WUs.
This's completely different than my 2k/xp experience which showed little to no difference. MY guess is that the win boxes have other junk running in the background hogging CPU time. I'd suggest checking taskmangler for junk taking a share of CPU time, or checking each boxes private summary page to see what percent of the time boinc's getting the processor.
Hello Just been reading
)
Hello
Just been reading this thread - haven't been here for a while. I have tweaked XP as far as I could, turning services off, registry tweaks etc, and have been running a dual-boot with assorted flavours of Linux. I don't have exact figures but if memory serves me correctly (??) using the old WUs from Einstein gave me the following results:
on 2.6 Intel with 768 meg RAM
Windows XP (service pack 2 added after install): around 8 hours 10 minutes
Kubuntu (Ubuntu with KDE GUI): around 7 hours 30 minutes
DSLinux: around 7 hours 15 minutes
SUSE 9.1 Linux: around 7 hours 45 minutes
(NB: I have reason to believe from reviews that I have read that SUSE 10.1 when it is released, will be somewhat faster than 9.1, tho' by how much I don't know)
I am running Gentoo Linux at the moment dual-booting Windows XP, but cannot comment on th WU times now as the changeover to assorted sizes happened before I installed it.
Hope this is of some use
Gray
RE: Hello Just been
)
Just to add more detail: in Windows I was running Zonealarm firewall and NOD32 antivirus in the days when those figure were applicable. An antivirus obviously affects computational power, and Zonealarm, while VERY neccessary, unless behind a router, is also a substantial drag on performance.
KDE in linux is defintely getting faster, but a window manager as supplied by DSlinux - Fluxbox, I think - is far lighter in its cpu requirements. If I was to build a box just for BOINC in the future, I would probably use Gentoo Linux and do a GUI-free install. Why Gentoo ? Hehe, I'm a control freak - pure and simple.
Gray
cheers - Gray
Hey again. If there are
)
Hey again.
If there are processes stealing cpu time in XP, It is not of my doing. Each install is fresh out of the box with OS defaults and NO applications installed.
XP has a built in firewall that I will check the next time I am in the lab.
After this amount of time I am of the opinion that we could rule out different WU sizes as they should have averaged out by now.
The CPU efficiencies are vastly different machines and I was wondering just what that means and how to tweak it up a notch on the MS OS's. this statistic seems o be pretty relevant to the overall performance.
3 more machines are available for experimentation, please give suggestions as to which Linux versions would be best to compare. I am thinking of taking down Moe and Curly as the performance is just pathetic on a dedicated box and I think that the point has been proven and the test should be considered over. Once I tell them all to cease taking further WU's, I will compare the work done and make a conclusion when all of the WU's are verified. You are all welcome to look at the machines and compare.
Non GUI Linux...I would like to, but I may not have the technical skill as I am a noob. Perhaps that can be Experiment #3.
RE: Hey again. If there
)
Hello there
There are a number of sites on the web that give info and stuff like services and other tweaks for XP - you might want to have a look here: http://www.theeldergeek.com/ and do a search for services. You might want to turn off all the GUI effects as well if you are using the box for anything barring crunching. Should you need an antivirus, grab either AVG, which has a free version or NOD32 which looks to have a good virii locating ability, but costs money - both are comparatively low on system resources.
If you have a broadband connection such as adsl with a router (wireless is rather more tricky I believe), Gentoo provides a decent set of documentation which you can access as you install. You'd need to get a minimal cd from one of their mirrors. This is quite small and just figures your dhcp connection plus hardware detection. It also gives you a text browser called Links with the address to their online docs supplied on the screen once you are in their command-line interface. I'm possibly a slightly advanced Win XP user, but I'm typing this from the Gentoo side of a Dual Boot system so it ain't THAT hard. Be warned tho' READ THE MANUAL !! Don't make any assumptions ! The forums are good too there, tho' they expect you to at least have tried to find the info before asking... and you can access them through links. The only issue timewise is the kernel compilation (use Genkernel if a noob - safer I feel) as that takes around 45 minutes on my 2.6 box. The other compilations are small, unless you want to go for a GUI, where the big files start... (Oh and if you use "grub" as a bootloader between XP and Gentoo, make sure you setup gub in (hd0) NOT (hd0,0). The latter is the usual windows partition and I bolixed my system that way before - late night installs are not really a good idea....
Good Luck - Gray
Hello again If you don't
)
Hello again
If you don't want to go through the Gentoo install, have a look at Damn Small Linux - only around 50 megs, has a GUI with some basic apps. It is designed as a live CD, but can be installed to a linux partition - maybe grab a Gentoo minimal CD which will allow you to make partitions fdisk /dev/hda for usual single drive box, and mke2fs -j /dev/hda(partition number) (without brackets) to give you an ext3 file system (one linux version of for instance fat32 for windows). If you have a router it picks it up and I had BOINC running easily in the dsl normal users home folder.
Gray
Hello Just a slight update
)
Hello
Just a slight update - I installed Gentoo Linux around 2 weeks back and now have dropped a BOINC binary into the system. I have just finished getting the Albert application and associated files and have begun to crunch - will be interesting to see if there is any significant speedup - especially as I have not really made any attempt to keep the system small and lightweight - I'm looking for a windows replacement so it has to do everything that windows used to do.
Obviously there will not be an exact measurement as there was with Einstein, more's the pity, but still worth the try anyhow - quite aside from me helping science in the process
Gray
RE: Someone else stated
)
Found a WU. But I don't like the outcome :(
For the record. Both processors are AMD XP 2600+. Mine (#464612) running Linux did the WU in 5,739.24s. The other (#48772) running XP 4,529.41s.
Michael
Team Linux Users Everywhere
RE: Found a WU. But I don't
)
I think this difference comes from the counterclock signals. Probably the other cpu is overclocked. (XP2600+ factory speeds: 1917MHz,2080MHz,2083MHz,2133MHz)
From what I can tell, Windows
)
From what I can tell, Windows Clients still seem to be some 10-15% faster.
I got two almost identical Hosts for comparison :
AthlonXP 3000+, Win2000 SP4
AthlonXP 3000+, Mandrake Linux 2.6.3-7
The Results are consistently approx. 14000s (Win32) vs. 16000s (Linux)
(but still much better than the ~50-60% Linux disadvantage we saw last year with the older Clients)
It's indeed difficult to compare AMD CPU's with different Cores, as Einstein seems to care only for raw clockspeed (appears to fit easily into 256k L2 Cache or less).
Performance is also relatively lower on low clocked Athlon64 CPU's (e.g. 1800MHZ So939 Athlon64 3000+, which is easily slower than faster clocked AthlonXP's on the Einstein client but excels fine according its Performance Rating in other Projects)