Nqueens Push

[gopher_yarrowzoo]

Oh and Welcome Chuck , I'll show up in stats too soon, I hope!

[Chuck]

Hi Goph!
Hi All..... Just shot an email to Laurenu2 and decided to check in here. Laurenu2 just pushed us to 9th position. ( I have to learn where we stand & all )

Some things for linux (64 bit) users to watch out for... 6.4.5 does not work properly on Fedora AMD installs on ANY CPU. The CORRECT version is 6.2.15.... BUT this means you MUST VERIFY the 'network always available' in the LOCAL OVERRIDE SETTINGS... and that it is NOT doing anything by the preferences critial to ops. Otherwise you will have work waiting to report, but never reported. One machine had 700+ claimed credits waiting for the 'update', just since 5am.

As I told Lauren.... there was a ton of work stacked up ready to send in... It's now uploading/reporting & processing.... You should see things pick up now (I hope I got it all).

To add injury to insult.... I had an 'overcollection of farm dust' get into my Gigabyte coolers.... As such, the cpus shut down at 70+C like they should, but the nVidia graphics card sitting next to it could not handle the 70C and gave up the ghost. I will be off to the store this PM to see if anything reasonable (8800, 8800GT, etc is available).... I'ld take anything that can handle the heat in the event of another farm dust buildup.
I also lost 3 HD's in that... but they were circa 2004 vintage and the heat obviously did them in.. no big deal there.
After a good washing of alcohol and relubricating of bearings and fresh mounting on the cpus... all is good again... 39-42C at full load per CPU... including being overclocked by 6% until compound cures.

To get things running and 'up the heat'... I turned off the X-server, and am running the one machine on a (DO NOT LAUGH) STB 128 Velocity (8MB RAM) PCI-33 card. It works fine in text mode but not well in X.... go figure! (it's my 'dire emergency' card!)

As for the project... Now the settings are set.... All non-boinc work suspended to make up for the stupidity / frustration. (yes, I was an unhappy camper!)

Project Overall: Perhaps being redundant, but running in Text mode (./run_client) is much faster than bringing up the overhead of X and the BOINC GUI if settings are all KNOWN CORRECT and avoiding all the darn RPC calls BOINC makes.


Thanks again for the welcome Goph..... It's good to be up and running.... %#@!!$$ BOINC.... lol

Laurenu2 and all have done a great job pushing us up and I hope I can contribute to that now things are running as they should be. Four spare cpu coolers are in prep to be ready for the other machine, should another also need a 'de-farming'....

Hope my efforts help.


Chuck

[Chuck]

Laurenu2,

Kindly kick in the afterburners please!!!!

A number of my 'pending' credits are waiting for you!!!

Anything I can help that "poor Windows XP machine" with?



Chuck

[Bok]

actually that's nothing to do with v6.4.5 I don't think.

I was running 6.4.5 on a couple of 64bit AMD's under CentOS just fine, though they now are running 6.5 (and have issues due to scheduling and GPUGrid making it think it needs more cpu resource ) The boinc developers are working on this for the 6.6 series coming out probably in the next few weeks..

Waiting to report just means that it hasn't ran a scheduled report to the server in a while. This could be due to many reasons, but mainly it's the value of the 'connected to internet every X days' which should be set to 0 for always on, and yes this can be overrode at the client-side but it's not really necessary...

After all that though, if you can I'd stick with 6.2.19 or 6.2.21 anyway due to the scheduling problems which have cropped up recently.

Bok

[Chuck]

actually that's nothing to do with v6.4.5 I don't think.

I was running 6.4.5 on a couple of 64bit AMD's under CentOS just fine, though they now are running 6.5 (and have issues due to scheduling and GPUGrid making it think it needs more cpu resource ) The boinc developers are working on this for the 6.6 series coming out probably in the next few weeks..

Waiting to report just means that it hasn't ran a scheduled report to the server in a while. This could be due to many reasons, but mainly it's the value of the 'connected to internet every X days' which should be set to 0 for always on, and yes this can be overrode at the client-side but it's not really necessary...

After all that though, if you can I'd stick with 6.2.19 or 6.2.21 anyway due to the scheduling problems which have cropped up recently.

Bok

Bok,

You are probably correct that it has nothing to do with 6.4.5 specifically. I think it has more to do with the Network Manager & other noise now in FC-9 and FC-10 which got accidentally pushed back into FC-8.

I have experienced the scheduling problems (minor) with 6.2.15 on FC-8.... they are manageable/correctable.... I know 6.4.5 is good with other 64 bit linuxes, but I was specific to the Fedora AMD (FC-8, FC-9, and probably the FC-10 which is coming out).

I will try 6.2.19 or .21 if you think that's got a better scheduler..... What it reports to the host (that my client is only 16% active, yet runs 98-99% efficiency) doesn't make sense. The scheduling and interface of 6.4.5 on Fedora (redhat) just doesn't want to work right, regardless of what I do, but then the Network Manager and related VPN packages have SVN conflicts like crazy.... (to the point of not even installing!!!!) ... Somehow the Tree branches for FC-9 seem to have been pushed onto FC-8 and caused heretofore unseen problems in FC-8. One such example is a standard NFS mount..... Can you believe 'No route to host' ?? when the IP of the host is in the host table and you can both PING and SSH and everything else to it, including run Sun xVM and get a SMB share from it???

Is it worth the effort to give it to give 6.2.19/21 a try? Does it work 'better' ?


Thanks,
Chuck

[Bok]

Sounds like your problems are more with the OS than Boinc anyway right ?

I never use the packages for BOINC anyway, I just wget the tarball and untar it and then

./boinc --allow_gui_remote_rpc via screen (to let me utilize boincview)

Longtime since I used FC, I prefer CentOS these days for crunchers personally, mainly because I use it for my work stuff as my comany only supports Suse or RHEL (but CentOs works fine )

If you are getting 'no route to host' messages, well that's a different kettle of fish... I'd switch of ipv6 first then start checking the routing..

Bok

[Chuck]

Bok,
There is no 6.2.19 or 6.2.21 left anymore in the x86_64 side at boincdl.

I would much rather download the sources and compile my own. that way I get an AMD64 scheduled version anyway.... meaning the benchmarks come out correctly, etc.

I do have my client connection value == 0. It likes that.... Now, I found something, possibly related and please verify that if you specify 100.00 (make sure to put in the .00) for your cpu utilization and loading parameters (bottom two) of the processors tab that all is still ok for you; that you don't get a stderr report from the project attached to your results. It reports errors for me... and then 'ignores' the invalid XML value of 100.00 It also ignores the invalid date.... I presume the completion date deadline (in unix time format) because it's also a float (.00000) and not an integer (which time and cpu% should be..... as in eg.100%).

I will continue to look at results and see what I can find... and perhaps we can figure out if it is my machine/client or the project ??? It has not impacted my results validity OTHER than lowering the claimed credits and artificallly raising the CPU time used. ~10,000 seconds???? these things are done in approx 2 hrs except for the P4, which does have some issues and takes about 9,000 seconds

If this is 6.2.15 related, then I understand..... and will have to find a way to get something better... even if 32 bit mode only.


As for the NFS issue.... I know that is the 'NetworkManager' package..... I have a work around by putting in manual routing statements into the table for host->host NFS connections and all is fine. Yet if you look at 'netstat -r' everything is fine there.... Again, another FC-9 error that has NEVER existed before in FC-8. I never had this problem until FC-9 came out. Perhaps it's time to switch / try something else in spite of my long standing familiarity to Suse and Redhat Linux.
(More personal + professional with Redhat, but experience with Suse at the personal level as well in the early days)....

Is is perhaps time to drop BACK to a 32 bit install for running boinc? I know that's why my benchmarks are messed up.... it was a perfect piece of algebra to compute the correct value that BOINC reported.

It does make sense for 32 bit boinc..... if this project is 32 bit based (even if 64 bit wrapped)....

Things like OGR really do benefit from 64 bit mode. Personal projects (video, audio, math, etc) and true boinc 64 bit math are fantastic in 64 bit mode. aka... anything moving data or doing real 64 bit final or interim result math, benefits. (more work / clock... plain and simple).

Advice? A 32 bit and a 64 bit install of Fedora or Suse or ????? (just not Gentoo or Ubuntu please.... <grin>)


FWIW: BOINC is the ONLY application I have trouble with in 64 bit mode other than the OS NFS glitch in the routing table


C.




here is what I got....

<core_client_version>6.2.15</core_client_version>
<![CDATA[
<stderr_txt>
Unrecognized XML in parse_init_data_file: computation_deadline
Skipping: 1231497971.000000
Skipping: /computation_deadline
Unrecognized XML in GLOBAL_PREFS::parse_override: mod_time
Skipping: /mod_time
Unrecognized XML in GLOBAL_PREFS::parse_override: max_ncpus_pct
Skipping: 100.000000
Skipping: /max_ncpus_pct
APP: NQueen checkpoint done (6.667 %)
APP: NQueen checkpoint done (13.333 %)
APP: NQueen checkpoint done (20.000 %)
APP: NQueen checkpoint done (26.667 %)
APP: NQueen checkpoint done (33.333 %)
APP: NQueen checkpoint done (40.000 %)
APP: NQueen checkpoint done (46.667 %)
APP: NQueen checkpoint done (53.333 %)
APP: NQueen checkpoint done (60.000 %)
APP: NQueen checkpoint done (66.667 %)
APP: NQueen checkpoint done (73.333 %)
APP: NQueen checkpoint done (80.000 %)
APP: NQueen checkpoint done (86.667 %)
APP: NQueen checkpoint done (93.333 %)

</stderr_txt>
]]>

[Chuck]

Sounds like your problems are more with the OS than Boinc anyway right ? ........ If you are getting 'no route to host' messages, well that's a different kettle of fish... I'd switch of ipv6 first then start checking the routing..

Bok

I found it...... combo of a) firewall/ iptables and b) SELINUX...... i know the game now..Fedora's "permissive" is not as permissive as it claims, neither is the firewall. Ports are open, 'no route to host magically disappeared, proper ports are open / specified and all is


sorry for off-topic, but now I can control it all from one machine AND have the proper ports open..... DAMN, it's tighter than a ........... censored..............

C.

[LAURENU2]

Now I am at near full power I Cleaned out all the old work
But now I have found 2 burned out quad fuel injectors So Afterburners will not ignite.
I will try to pick up new ones tonight if I can