IRIX and Software

Seti Enhanced - Page 1

Although not IRIX related, I've come to the realization that Seti Enhanced sucks!!!

I have two windoze boxes crunching, a dual P3 1 GHz VW320 and a 2.8 GHz P4. The dual P3 was turning out a WU about every 10 to 12 hours per processor under normal seti, just checked since upgrading to enhanced and I see the current two WU's being run with CPU time elapsed 52 hours and time to completion 138 hours. The P4 was turning out a WU every 3 to 4 hours, that went to 12 to 16 hours. I've also noted lower granted credit for work done.

Anyone else seeing this?
-ks

:Onyx: :Onyx: :Crimson: :O2000: :Onyx2: :Fuel: :Octane: :Octane2: :PI: :Indigo: :Indigo: :O2: :O2: :Indigo2: :Indigo2: :Indigo2IMP: :Indy: :320: :540: :O3x0: :1600SW: :1600SW: :hpserv:

See them all >here<
This is okay. The new client is searching much more precisely and therefore needs more cpu-time. Running MacOSX or HP-UX, it's the same.
Image
Ok, since their not sending out any regular work units anymore, and my SGI's are sitting idle again, anyone have an Enhanced client working??
-ks

:Onyx: :Onyx: :Crimson: :O2000: :Onyx2: :Fuel: :Octane: :Octane2: :PI: :Indigo: :Indigo: :O2: :O2: :Indigo2: :Indigo2: :Indigo2IMP: :Indy: :320: :540: :O3x0: :1600SW: :1600SW: :hpserv:

See them all >here<
kshuff wrote: Ok, since their not sending out any regular work units anymore, and my SGI's are sitting idle again, anyone have an Enhanced client working??


i tried to compile it some time ago but the code really sucked :D
r-a-c.de
foetz wrote:
kshuff wrote: Ok, since their not sending out any regular work units anymore, and my SGI's are sitting idle again, anyone have an Enhanced client working??


i tried to compile it some time ago but the code really sucked :D


Yeah, I've read in another thread. Thought maybe someone had succeeded
-ks

:Onyx: :Onyx: :Crimson: :O2000: :Onyx2: :Fuel: :Octane: :Octane2: :PI: :Indigo: :Indigo: :O2: :O2: :Indigo2: :Indigo2: :Indigo2IMP: :Indy: :320: :540: :O3x0: :1600SW: :1600SW: :hpserv:

See them all >here<
Well, guys, let's face it: Irix is out of the game as far as distributed computing is concerned. Rest in peace ! :cry:
Image
ulenz wrote: Well, guys, let's face it: Irix is out of the game as far as distributed computing is concerned. Rest in peace ! :cry:


Maybe not, I posted to the message board and it was turned over to the IRIX porter. Now we play the waiting game...
-ks

:Onyx: :Onyx: :Crimson: :O2000: :Onyx2: :Fuel: :Octane: :Octane2: :PI: :Indigo: :Indigo: :O2: :O2: :Indigo2: :Indigo2: :Indigo2IMP: :Indy: :320: :540: :O3x0: :1600SW: :1600SW: :hpserv:

See them all >here<
ulenz wrote: Well, guys, let's face it: Irix is out of the game as far as distributed computing is concerned. Rest in peace ! :cry:


no reason for making crappy code anyway.
r-a-c.de
ulenz wrote: Well, guys, let's face it: Irix is out of the game as far as distributed computing is concerned. Rest in peace ! :cry:
Not in my office... Running a 4-way distributed NASTRAN job as we speak :-)
What is NASTRAN ?
Image
ulenz wrote: What is NASTRAN ?
An engineering application that requires heavy number crunching...
Hi guys,

Well, I managed to compile the BOINC client (in the process adding the code to correctly identify the RAM, swap space, number of CPUs) and the [email protected] enhanced client.
I haven't had the time to package this properly and the binaries have some dependencies they shouldn't right now, but if you follow the instructions on my download page, they should work until I get around to creating a proper tardist.
Some info on the way the binaries were compiled:
1. BOINC client gcc/g++ 3.2 (the Nekoware gcc 3.4 lacks libstdc++ which is required). No real optimisation, but it isn't required for this binary.
2. [email protected] client - gcc/g++ 3.2 - Unfortunately, porting this to MipsPro (I have only 7.3.1.3) would be an exercise in frustration, however,I have tried to build it as optimised as possible (i.e. -O3 -mips4 -mabi=n32 -mfused-madd -fomit-frame-pointer -funroll-loops).
In adiition to this - the [email protected] binary is now linked (statically) against libfftw (3.1.1 or above) and it is within this library that a lot of the computation takes place.
Fortunately, this library was written in C and by people who had cross-platform portability in mind, so it compiles well using MipsPro. Incidentally, the seti porting guys have been playing with the lib more recently, and it is already showing signs of becoming less portable. I compiled the original. I doubt that any changes made would improve performance on MIPS chips, but I will test this eventually.
The library was compiled with:
-O3 -mips4 -IPA -TARG:madd=ON:platform=ip30 -OPT:Olimit=0, i.e. fairly aggressively.
:)
Note: I didn't use -Ofast as I thought that it may have too much of an impact on floating point values.
The library was compiled in single precision mode - gives approximately 4% speed increase over standard (double precision) mode.
Hmm, I wonder whether performance would be better if calls to FFTW were replaced with the SCSL implementation. Unfortunately, I won't have the time to try this.

Anyway, the current archive is here:
http://nikv.pwp.blueyonder.co.uk/

The directory structure and the app_info.xml file are already set up for [email protected] enhanced. All you need to do (once you have installed the pre-reqs and the LD_LIBRARY_PATH has been set up) is to run 'boinc_client' giving the seti project url and your account key. Run boinc_client --help for instructions.
Boinc now includes 'boinc_cmd' binary which lets you check on the progress of the calculation. 'boinc_cmd --help' for more instructions.

Anyway, I am planning to package this up as a tardist in the next few weeks.
In the meantime, enjoy and let me know if it works for you.

Nik.
I have a much more serious and important question.

How come it says Team Picard but the topbanner on the site shows the retro Kirk Enterprise?
Maverick 3: Athlon X2 7750, 2gb, Windows Vista
Frank Dux: SGI Octane2 R12k 400mhz 1.5gb

"Chief, look! I learned to make fire! Who knows what we could do with this... We should learn to control it!"

"Ridiculous. How can you justify wasting time and effort on this so-called 'fire' when our children are freezing to death at night?"
nvukovlj wrote: Hi guys,

Well, I managed to compile the BOINC client (in the process adding the code to correctly identify the RAM, swap space, number of CPUs) and the [email protected] enhanced client.
I haven't had the time to package this properly and the binaries have some dependencies they shouldn't right now, but if you follow the instructions on my download page, they should work until I get around to creating a proper tardist.
Some info on the way the binaries were compiled:
1. BOINC client gcc/g++ 3.2 (the Nekoware gcc 3.4 lacks libstdc++ which is required). No real optimisation, but it isn't required for this binary.
2. [email protected] client - gcc/g++ 3.2 - Unfortunately, porting this to MipsPro (I have only 7.3.1.3) would be an exercise in frustration, however,I have tried to build it as optimised as possible (i.e. -O3 -mips4 -mabi=n32 -mfused-madd -fomit-frame-pointer -funroll-loops).
In adiition to this - the [email protected] binary is now linked (statically) against libfftw (3.1.1 or above) and it is within this library that a lot of the computation takes place.
Fortunately, this library was written in C and by people who had cross-platform portability in mind, so it compiles well using MipsPro. Incidentally, the seti porting guys have been playing with the lib more recently, and it is already showing signs of becoming less portable. I compiled the original. I doubt that any changes made would improve performance on MIPS chips, but I will test this eventually.
The library was compiled with:
-O3 -mips4 -IPA -TARG:madd=ON:platform=ip30 -OPT:Olimit=0, i.e. fairly aggressively.
:)
Note: I didn't use -Ofast as I thought that it may have too much of an impact on floating point values.
The library was compiled in single precision mode - gives approximately 4% speed increase over standard (double precision) mode.
Hmm, I wonder whether performance would be better if calls to FFTW were replaced with the SCSL implementation. Unfortunately, I won't have the time to try this.

Anyway, the current archive is here:
http://nikv.pwp.blueyonder.co.uk/

The directory structure and the app_info.xml file are already set up for [email protected] enhanced. All you need to do (once you have installed the pre-reqs and the LD_LIBRARY_PATH has been set up) is to run 'boinc_client' giving the seti project url and your account key. Run boinc_client --help for instructions.
Boinc now includes 'boinc_cmd' binary which lets you check on the progress of the calculation. 'boinc_cmd --help' for more instructions.

Anyway, I am planning to package this up as a tardist in the next few weeks.
In the meantime, enjoy and let me know if it works for you.

Nik.


Way cool, thanks Nik. I'll download tomorrow and let you know how I make out :D
-ks

:Onyx: :Onyx: :Crimson: :O2000: :Onyx2: :Fuel: :Octane: :Octane2: :PI: :Indigo: :Indigo: :O2: :O2: :Indigo2: :Indigo2: :Indigo2IMP: :Indy: :320: :540: :O3x0: :1600SW: :1600SW: :hpserv:

See them all >here<
One thing I forgot to mention - the gcc libraries (stdc++, gcc_lib) will currently have to go into /usr/local/lib. Sorry about that.
My plan is to compile gcc 3.4.6 and upload that to Nekoware and then compile these binaries using that version of gcc/g++ so that a proper pre-req. order can be created.

Nik.
nvukovlj wrote:
kshuff wrote:
Hey Nik, have you had a chance to package this into a tardist yet?


Hi,

I will try to do it this weekend. Didn't get the chance to do it in the past week.

Nik.


You have the same problem as me, too many projects, not enough time :)
-ks

:Onyx: :Onyx: :Crimson: :O2000: :Onyx2: :Fuel: :Octane: :Octane2: :PI: :Indigo: :Indigo: :O2: :O2: :Indigo2: :Indigo2: :Indigo2IMP: :Indy: :320: :540: :O3x0: :1600SW: :1600SW: :hpserv:

See them all >here<
ulenz wrote: What is NASTRAN ?


Nasa Structural Analysis System, whatever that shit might do!?

Maybe something like finite element crap.
Yes the works.... but as previous boinc version it works like a blackhole. Dunno how seti work internaly. As reportet earlier the project have bandwidht problems and not enough WU for all clients.

Also a big problem is coming up after an outage. Must be the same like a Ddos attack :).

To avoid this the client sets a long timeout period befor start with reconecting. If you take a look into one of the XML file you can see a unixtimestamp value when the client connect next time to the network. I have to wait more than 24h :)

After this time my client starts download WUs. But running the client like in the pass doesnt work so good anymore.

Code: Select all

nohup npri -w ./boinc_client >> boinc.log &


- A lot of seti tasks shown as sleeping status when using top, but 99% cpu
- Sometime there is no activity. If i restart boinc the seti starts resuming
- Sometime boinc went away


I have no clue about these warnings

Code: Select all

2006-06-21 18:46:41 [---] Process 418475 not found

2006-06-21 13:43:57 [---] Using earliest-deadline-first scheduling because computer is overcommitted.
2006-06-21 13:43:57 [---] Suspending work fetch because computer is overcommitted.
2006-06-22 21:12:46 [[email protected]] Resuming task 01ap99ab.20150.22944.390890.3.10_0 using setiathome_enhanced version 512
2006-06-22 21:12:46 [[email protected]] Deferring task 01ap99ab.20150.22944.390890.3.13_1


I also attached the Einstein project to this boinc client. But it looks they have updated the client and older einstein versions didnt get any work and there is no source of the project.

Code: Select all

[o2k]:/usr/people/beh/krc $ pstree -p 488824
-+- 00001 root /etc/init
\-+- 488709 root screen -R boinc
\-+- 488723 beh /usr/nekoware/bin/bash
\-+- 488824 beh ./boinc_client
|--- 513375 beh setiathome-5.12.mips-sgi-irix
|--- 514759 beh setiathome-5.12.mips-sgi-irix
|--- 519186 beh setiathome-5.12.mips-sgi-irix
|--- 502245 beh setiathome-5.12.mips-sgi-irix
|--- 526817 beh setiathome-5.12.mips-sgi-irix
|--- 509293 beh setiathome-5.12.mips-sgi-irix
|--- 509758 beh setiathome-5.12.mips-sgi-irix
|--- 533112 beh setiathome-5.12.mips-sgi-irix
|--- 488815 beh setiathome-5.12.mips-sgi-irix
|--- 488830 beh setiathome-5.12.mips-sgi-irix
|--- 488834 beh setiathome-5.12.mips-sgi-irix
|--- 488835 beh setiathome-5.12.mips-sgi-irix
|--- 488837 beh setiathome-5.12.mips-sgi-irix
|--- 488842 beh setiathome-5.12.mips-sgi-irix
|--- 488845 beh setiathome-5.12.mips-sgi-irix
|--- 488849 beh setiathome-5.12.mips-sgi-irix
|--- 488851 beh setiathome-5.12.mips-sgi-irix
|--- 488854 beh setiathome-5.12.mips-sgi-irix
|--- 488861 beh setiathome-5.12.mips-sgi-irix
\--- 488864 beh setiathome-5.12.mips-sgi-irix


As a notice. It looks like that it needs more time now to compute a WU.


In the seti Forum i have read something about a reference WU which can be used for benchmarking.

@nvukovlj
can you provide your patched how getting boinc/seti compiled? I would like to take a look of the use of non nekoware libs like zlib,libjpeg. If i can work it out i would like to create a neko_boinc package.

regards
Joerg
Tried a few of your suggestions.

I am assuming that you are using the app_info.xml I included in the archive.

Yes I am.

It is possible that the seti client can't start up. Have you tried starting the seti client from the same shell you have used to start the boinc client ?

Tried running Seti by itself and yes you are right it does not start like it should. I get this error even after verifying Neko openssl is loaded and adding a setenv library path variable to its location.

Code: Select all

Octane 5% ./setiathome-5.12.mips-sgi-irix
15143:./setiathome-5.12.mips-sgi-irix: rld: Warning: Version search suppressed in ./setiathome-5.12.mips-sgi-irix because version (sgi0.9.7) of object libssl.so in liblist is not an sgi interface version.
15143:./setiathome-5.12.mips-sgi-irix: rld: Fatal Error: Cannot Successfully map soname 'libssl.so' under any of the filenames /usr/lib32/libssl.so:/usr/lib32/internal/libssl.so:/lib32/libssl.so:/opt/lib32/libssl.so:


Thanks for your help.

JR
Well, I don't have any problems running [email protected] and [email protected] on my PowerMac G5 and MacOSX.
Let's face it, guys:
The crunching power of older SGI-MIPS-cpus is poor in comparison with actual AMD-/Intel-cpus. Moreover, there are only a few machines available. Therefore it doesn't make much sense any longer to develop Irix-clients for distributed computing. Too much electricity consumption and too less crunching power. :(
Image