Badly optimized Win app ???

tapir
tapir
Joined: 19 Mar 05
Posts: 23
Credit: 462,935,446
RAC: 0
Topic 193384

Akosf Quad CPU Q6600 in "top 10" must be able to work at 6 GHz to reach that speed of processing, or he succeed to optimize win. app. ???

akosf Q6600@??? GHz took 13000 sec for 236.7 credit;
mine QX6700@3,5 GHz took 25000 sec for 237.3 credit

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 693,435,718
RAC: 96,079

Badly optimized Win app ???

Quote:

Akosf Quad CPU Q6600 in "top 10" must be able to work at 6 GHz to reach that speed of processing, or he succeed to optimize win. app. ???

akosf Q6600@??? GHz took 13000 sec for 236.7 credit;
mine QX6700@3,5 GHz took 25000 sec for 237.3 credit

It is no secret that Akos made crucial contributions to the SSE vector code and some optimizations of some trigonometric calculations that are already partly used in the latest Mac OS/Intel beta and the newest Linux/Intel power-user app.

http://www.newscientisttech.com/article/dn9180-programmer-speeds-search-for-gravitational-waves.html
http://einstein.phys.uwm.edu/contributors.php

While Akos was indeed looking for some water-cooling stuff recently, I would not be surprised at all if he's in addition also experimenting with SSE vector optimizations under Windows :-)

CU
Bikeman

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282,700
RAC: 0

Well, given what Bernd told

Well, given what Bernd told me when I asked, I think this DESERVES an explanation...

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282,700
RAC: 0

Here is the current OS

Here is the current OS distribution for hosts working on Einstein (per BOINCStats):

With the estimated time to finish S5R3 currently standing at 443 days, is there not incentive to finish the run sooner so that post-processing can begin?

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3,522
Credit: 693,435,718
RAC: 96,079

RE: Here is the current OS

Message 76485 in response to message 76484

Quote:

Here is the current OS distribution for hosts working on Einstein (per BOINCStats):

With the estimated time to finish S5R3 currently standing at 443 days, is there not incentive to finish the run sooner so that post-processing can begin?

Sure, but you can consider whatever Akos is using as experimental software, pre-beta if you want. If it turns out to be stable, cross-validating and compilable from a common source code basis, it will become official. This takes time and effort. If you can complte S5R3 in half the time but in the end are not sure whether you might have missed somethingm, you didn't gain anything.

Bikeman

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282,700
RAC: 0

RE: RE: Here is the

Message 76486 in response to message 76485

Quote:
Quote:

Here is the current OS distribution for hosts working on Einstein (per BOINCStats):

With the estimated time to finish S5R3 currently standing at 443 days, is there not incentive to finish the run sooner so that post-processing can begin?

Sure, but you can consider whatever Akos is using as experimental software, pre-beta if you want. If it turns out to be stable, cross-validating and compilable from a common source code basis, it will become official. This takes time and effort. If you can complte S5R3 in half the time but in the end are not sure whether you might have missed somethingm, you didn't gain anything.

Well, here is what was said, verbatim:

  • * "fix the fast 'linear' SIN/COS calculation code for all platforms. Currently this only works when compiled with Apple's version of gcc on MacOS Intel. In principle it should be generic and thus give soem speedup on all platforms."
    * "Currently the SSE code can only be compiled with gcc; and when linking gcc code to MSC code the debugging information of the gcc objects get lost. Therefore I find it rather unlikely that there will be a Windows App with an SSE 'hot-loop', say, this year."

So, while I appreciate the idea that perhaps something might be "pre-beta", the fact remains that certain statements fly in the face of what can be observed.

It would be GREATLY appreciated if a simple explanation of what's going on with that particular host could be made by Akos and/or Bernd.

Yes, I am a bit miffed. IMO, other Windows OS folks should be too.

Consider this another "whammy" on top of the other two already encountered (current results grabbed at maximum runtime variation cycle and the advantages held by AMD/Linux owners)...

FalconFly
FalconFly
Joined: 16 Feb 05
Posts: 191
Credit: 15,650,710
RAC: 0

Just remember that there was

Message 76487 in response to message 76486

Just remember that there was a time where things were not only opposite (Linux Hosts losing darely both on runtime but also claimed Credits) but also more extremely geared towards Win32 performance.

What we now have is a comparably small performance difference, nothing to get 'miffed' about.

And about Akos hosts, never use them as a reference.
If you're jealous, just remember he was the driving force behind getting hot-loops and SIMD into Einstein Applications in the first place (as far as I recall).

And to put things into perspective, you're inquiring about one of 427000 Hosts currently running for EAH. If you look around other Projects, one can always find unreasonably fast Hosts for their given CPU ID. It's normal (especially if they belong to Akos ;) ) and doesn't warrant public questioning.

(when I participated Alpha Testing for Akos EAH Application optimizations a long time ago, I saw performance increases of >75% myself, so seeing him running an exceptionally fast host is of no surprise)

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282,700
RAC: 0

RE: Just remember that

Message 76488 in response to message 76487

Quote:
Just remember that there was a time where things were not only opposite (Linux Hosts losing darely both on runtime but also claimed Credits) but also more extremely geared towards Win32 performance.

If I did not mention it at the time, then that would be my fault, as I believe both situations are equally unfair.

I am in the process of going ahead and getting a VM set up. I found an X2 4800+ system running Linux to compare against. I cannot compare against your 4400+ system, as I believe you told me it was not overclocked. My benchmarks were faster than the 4800+ system, but yet they had similar runtimes to my system. Setting up the VM and giving it a go is the only real way to find out. I've decided that if there is more than 3% increase in performance, then that is significant. I don't know how much overhead there will be in the VM. If there is a 10% improvement (what I'm expecting), then that is major.

If none of this pans out, then I, and others, are more the wiser and I will offer an apology and shut up.

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282,700
RAC: 0

RE: I am in the process of

Message 76489 in response to message 76488

Quote:

I am in the process of going ahead and getting a VM set up.

Geez!!! I really don't get why people think Linux is just the Bees Knees!

First attempt via the Add/Remove in Ubuntu 7.10 got me a "BOINC Manager" of version 5.10.8. OK. So it's not the newest, but it is newer than 5.8.16 that I'm using with Windows. Alright. Fire it up. DOH! It "can't talk to a client". WHAT?!?!?!?!?!?! The installation in Windows brings the client along with the manager...

Anyway, I'm dredging through oodles of web pages about sudo apt get and apt install and whatnot, but then I see this thing about not running as root, which Bernd mentioned as well, and there's this HUGE procedure about installing BOINC in Ubuntu that looks to perhaps take like an hour to wade through...

I'm going to go eat supper and contemplate (love that word, a jerk I know once used it towards me in their moment of power over me) my next actions...

Donald A. Tevault
Donald A. Tevault
Joined: 17 Feb 06
Posts: 439
Credit: 73,516,529
RAC: 0

RE: RE: I am in the

Message 76490 in response to message 76489

Quote:
Quote:

I am in the process of going ahead and getting a VM set up.

Geez!!! I really don't get why people think Linux is just the Bees Knees!

First attempt via the Add/Remove in Ubuntu 7.10 got me a "BOINC Manager" of version 5.10.8. OK. So it's not the newest, but it is newer than 5.8.16 that I'm using with Windows. Alright. Fire it up. DOH! It "can't talk to a client". WHAT?!?!?!?!?!?! The installation in Windows brings the client along with the manager...

Anyway, I'm dredging through oodles of web pages about sudo apt get and apt install and whatnot, but then I see this thing about not running as root, which Bernd mentioned as well, and there's this HUGE procedure about installing BOINC in Ubuntu that looks to perhaps take like an hour to wade through...

I'm going to go eat supper and contemplate (love that word, a jerk I know once used it towards me in their moment of power over me) my next actions...

Actually, your best bet is to download the BOINC client script from the BOINC website into your home directory. Then, enter "chmod a+x boinc_5.10.28_i686-pc-linux-gnu.sh" to set the executable permission on the script, and then enter "./boinc_5.10.28_i686-pc-linux-gnu.sh" to perform the installation. When you want to run BOINC, just enter "cd BOINC", and then enter "./run_manager".

Doing it this way solves five problems.

1) You're not at the mercy of the Ubuntu package maintainers to always have an up-to-date BOINC client.

2) You won't need "root" privileges to either run or install it.

3) This also makes the project file directory easier to find.

4) You don't have to spend time trying to figure out how to use Aptitude, Synaptic, or apt-get. (Well, eventually you will, if you become a Penguin convert. But, for now, there's no need for that.)

5) Client and manager both get installed at the same time this way, so you won't get that pesky "can't talk to a client" error.

When a new BOINC client gets released, just repeat the installation process with the new installation script. The old client will get upgraded, your settings and project files will remain intact.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,870
Credit: 115,692,606,749
RAC: 34,674,170

RE: Geez!!! I really

Message 76492 in response to message 76489

Quote:

Geez!!! I really don't get why people think Linux is just the Bees Knees!

The key thing to remember is that the two systems are simply ... different!

To some extent you have to abandon Windows power user techniques and learn a new set. After a certain amount of pain and suffering the end result is quite rewarding.

I'd just like to endorse what Donald has said about how to install/upgrade BOINC on a Linux system. It really is easier to install and upgrade BOINC on Linux. I'd just like to mention a couple of very minor differences and a couple of small enhancements to what he wrote. I'm starting with the assumption that you have a Linux installation with a GUI (eg KDE) and a non-root user account you wish to use for BOINC.

As Donald mentions, download the appropriate BOINC installation script to your home directory. Whilst logged in as your non-root user, simply feed the shell script to a shell by issuing the command (no need to change any permissions):-

sh boinc_ubuntu_5.10.28_i686-pc-linux-gnu.sh

This will create a BOINC subdirectory if one doesn't already exist or will add the new files if one already exists.

As I prefer to run BOINC as a daemon, I don't bother with the runmanager script. I find a local startup script (in my case it's /etc/rc.d/rc.local) and I add the following lines:-

if [ -x /home/gary/BOINC/boinc ]
then
echo "Starting BOINC in 20 seconds ..."
( cd /home/gary/BOINC; sleep 20; sudo -u gary ./boinc --daemon --allow_remote_gui_rpc ) &
fi

You will notice a 20 second delay built in. On my Linux systems during startup, the network takes a bit to be fully up and BOINC complains if the network isn't up when it's trying to startup itself.

So BOINC always starts automatically whenever the system boots. Quite often I need to stop and restart BOINC so I create two extra scripts and put links to them on the desktop. Being very imaginative, I call then StartBoinc and StopBoinc respectively:-

if [ -x /home/gary/BOINC/boinc ]
then
( cd /home/gary/BOINC; ./boinc --daemon --allow_remote_gui_rpc )
fi

cd /home/gary/BOINC
./boinc_cmd --quit

This means that I can start or stop BOINC directly from the desktop any time I like with a single click on the appropriate icon.

Hope this helps.

Cheers,
Gary.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.