| Author | Message |
|
|
|
using command line client on Debian Etch amd64 (remote server):
16:15 0 rapture /etc/boinc-client # boinc_cmd --project http://abcathome.com/ update
16:19 1 rapture /etc/boinc-client # boinc_cmd --get_messages 0
[...]
ABC@home 1 1201101515 Sending scheduler request to http://abcathome.com/abc_cgi/cgi
ABC@home 1 1201101515 Reason: Requested by user
ABC@home 1 1201101515 Requesting 8640 seconds of new work
ABC@home 1 1201101521 Scheduler request succeeded
ABC@home 1 1201101521 Message from server: platform '' not found
16:19 0 rapture /etc/boinc-client # uname -a
Linux rapture 2.6.18-5-amd64 #1 SMP Sat Dec 22 20:43:59 UTC 2007 x86_64 GNU/Linux
what gives? |
|
|
|
|
|
The platform '' message indicates that your machine is not set up such that ABC is able to recognize what OS is running in order that it can decide on the work to send like Linux, OSX, Windows. Most probably if you attach to any other project you get the same response until that's fixed. Somewhere (not at ABC) read how to make FreeBSD identify itself as Linux with proper bit size, so the server would send that.
____________
Coelum Non Animum Mutant, Qui Trans Mare Currunt |
|
|
|
|
|
client installed straight from debian repos.
/etc/boinc-client/client_state.xml says:
<platform_name>x86_64-pc-linux-gnu</platform_name>
what else could it need? i've searched for this error on both boinc wiki and forums, to no avail. |
|
|
|
|
client installed straight from debian repos.
/etc/boinc-client/client_state.xml says:
<platform_name>x86_64-pc-linux-gnu</platform_name>
My client_state.xml also says
<platform_name>x86_64-pc-linux-gnu</platform_name>
and it works pretty good on both of my 64-bit Fedora boxes. I seriously doubt Debian or any other Linux needs to be identified as FreeBSD. I believe Sekerob is erroneously recalling some posts that tell how you can get some projects' Linux apps to download and run on FreeBSD.
what else could it need?
Having the correct platform_name in client_state.xml is just the first step. Next, your system needs to send the platform_name to the server in a scheduler request. You might be able to check see if that is happening. Exit BOINC and then check sched_request_abcathome.com.xml in your boinc dir and see if it has a line like <platform_name>x86_64-pc-linux-gnu</platform_name> too. It should. Mine does.
My hunch is that when your boinc.exe builds the sched_request_abcathome.com.xml it doesn't parse the platform name from client_state.xml properly (maybe a syntax error in the xml?). Instead of catching the error it just puts "" or some other nonsense instead of x86_64-pc-linux-gnu. Just a hunch but if it seems to hold water and you feel like driving yourself nuts then maybe hand trace client_state.xml and look for a missing/extraneous <, >, / or whatever. There may be more than 1 such error in client_state.xml. You can reduce the number of lines you need to search through by draining your cache of work from other projects. Also, errors will be easier to spot if you use viewer that has syntax highlighting, eg. Firefox. Code editors have even better syntax highlighting than Firefox.
Or maybe client_state.xml isn't boogered, maybe you have a buggy boinc.exe from Debian repos in which case grabbing the latest recommended version from Berkeley might fix it.
I would say if you can download work from other projects then the problem is probably not boinc.exe and more likely a bad client_state.xml.
|
|
|
|
|
Having the correct platform_name in client_state.xml is just the first step.
actually it sounds like boinc/debian bug, both or either. i'm not sure as i'm no expert in boinc and it's rather poorly documented.
it appears that the client actually ignores /etc/boinc-client/client_state.xml. on (re)start, it generates /var/lib/binc-client/client_state.xml and in that one <platform_name></platform_name> is empty. even tried ln -s /etc/boinc-client/client_state.xml /var/lib/binc-client/ to no avail. it removes the symlink and generates wrong again.
i think i'll pass. the whole idea was to donate some spare cpu cycles to the advancement of science with minimum fuss. it's bad enough that abc@home is using a closed source binary, which is doing god-knows-what. i'm not gonna risk putting it on a production server. |
|
|
|
|
Having the correct platform_name in client_state.xml is just the first step.
actually it sounds like boinc/debian bug, both or either. i'm not sure as i'm no expert in boinc and it's rather poorly documented.
it appears that the client actually ignores /etc/boinc-client/client_state.xml. on (re)start, it generates /var/lib/binc-client/client_state.xml and in that one <platform_name></platform_name> is empty. even tried ln -s /etc/boinc-client/client_state.xml /var/lib/binc-client/ to no avail. it removes the symlink and generates wrong again.
The stock boinc executable generates all the xml and various files it uses in the same directory it resides in. It looks like your boinc executable has been altered to keep certain file types in certain directories to make it "easier" to organize and find things. Looks like they generated 2 client_state.xml in 2 different places. The one has the platform name written in it but the other doesn't. I bet when it's time to build the scheduler_request they read the platform name from the client_state.xml that does NOT have the platform name written.
I'm no Linux expert and my choice is Fedora rather than Debian but... I bet if you edit /var/lib/binc-client/client_state.xml and put the platform name in there it will all work properly. On the other hand there may be other bugs. Still it's worth a try.
If that doesn't work then uninstall the package and download the standard package from Berkeley. There's no installer, you just extract the files from the archive, put them in whatever directory you like. You can fire it up manually just by clicking the manager icon which starts the client if it's not already running. There are other easy options for a manual start, none of which are terribly inconvenient on a machine that runs 24/7. Or you can install it as a service though the docs I've read for that are not the best.
i think i'll pass. the whole idea was to donate some spare cpu cycles to the advancement of science with minimum fuss. it's bad enough that abc@home is using a closed source binary, which is doing god-knows-what. i'm not gonna risk putting it on a production server.
Well, never gamble what you can't afford to lose but you seemed to think it was a good idea at one time. If it's just the problems and frustration you've encountered so far then give it another think and keep asking questions... plenty of noobs with less Linux experience than you have made it work with not too much trouble and you can too. Plenty of help available here, just ask.
|
|
|
|
|
The stock boinc executable generates all the xml and various files it uses in the same directory it resides in. It looks like your boinc executable has been altered to keep certain file types in certain directories to make it "easier" to organize and find things.
it is customary to keep all system wide config files in /etc/. this has many good reasons. boinc keeps it's configs in ./, that's why debian attempted to correct it by moving (what they thought) config files to /etc/boinc-client/ and placing symlinks in /var/lib/boinc-client/.
Looks like they generated 2 client_state.xml in 2 different places. The one has the platform name written in it but the other doesn't. I bet when it's time to build the scheduler_request they read the platform name from the client_state.xml that does NOT have the platform name written.
I'm no Linux expert and my choice is Fedora rather than Debian but... I bet if you edit /var/lib/binc-client/client_state.xml and put the platform name in there it will all work properly. On the other hand there may be other bugs. Still it's worth a try.
doesn't work. it appears that when boinc is (re)started it unlink()'s /var/lib/binc-client/client_state.xml and generates it from scratch.
If that doesn't work then uninstall the package and download the standard package from Berkeley.
server administration handbook, page 1, paragraph 1: thou shall not install random packages found somewhere on the internet on thy server ;^)
There's no installer, you just extract the files from the archive, put them in whatever directory you like. You can fire it up manually just by clicking the manager icon which starts the client if it's not already running. There are other easy options for a manual start, none of which are terribly inconvenient on a machine that runs 24/7. Or you can install it as a service though the docs I've read for that are not the best.
you have obviously never administered a remote server ;). there's no X, no gui, no icons to click. at best, there's some kind of web interface.
i think i'll pass. the whole idea was to donate some spare cpu cycles to the advancement of science with minimum fuss. it's bad enough that abc@home is using a closed source binary, which is doing god-knows-what. i'm not gonna risk putting it on a production server.
Well, never gamble what you can't afford to lose but you seemed to think it was a good idea at one time. If it's just the problems and frustration you've encountered so far then give it another think and keep asking questions... plenty of noobs with less Linux experience than you have made it work with not too much trouble and you can too. Plenty of help available here, just ask.
i thought it might be a good idea because (a) there are servers with plenty of spare cpu cycles and they are all running 64bit linux; (b) abc@home greatly benefits from 64bit platforms, and (c) on gentoo it just works(tm). ok, i had to email abc people asking how to register an account using command line, but their reply was quick and very helpful.
on debian, what should have been a 5-minute job, became several hours. and i don't think debian is to be blamed, rather boinc and abc developers.
for anyone interested, the problem was resolved after upgrading boinc-client to 5.4.11-4 from debian backports <http://www.backports.org/dokuwiki/doku.php?id=instructions>.
for the abc@home project manager(s):
if you are interested in attracting cpu time from linux servers, you have to make sure that:
1) your project works out of the box on boinc cluents included in major server distros (the ones with long term support). this means rhel and it's clones (centos, white box, etc), novell/suse, debian stable and ubuntu LTS.
2) minimum system requirements, including version of boinc client supported, are clearly posted on the front page.
3) (link to) instructions on how to jump start the project using command line client clearly posted on the the front page.
4) open source abc app for security audit. |
|
|
|
|
|
Hi,
Ignoring that comment where Dagorith read my note as making Linux look like FreeBSD, where the example was to e.g. make FreeBSD 'look' like Linux, you found the answer and in fact found the same on a German website proposing to try a newer distro. http://www.boinc-team.de/portal/showtopic.php?threadid=1277
Problem solved. It would have been nice for U as the thread-opener to be able to insert [RESOLVED] in the title. Helps people to quickly identify which threads contain 'The' answer. The standard BOINC forum software is not great and ABC is update(s) behind on that (still not permitting OP title editing).
You might when things have settled consider going to a 5.10.xx release. There's also now an official Berkeley port to 64 bits on the topic of source credence. There's lots of features there that more projects start to use like on the fly up/download file compression. Another nice one is the client checking if buffered work was marked as redundant or faulty by the server admin. Saves bandwidth and crunchtime.
Take care
____________
Coelum Non Animum Mutant, Qui Trans Mare Currunt |
|
|
|
|
you have obviously never administered a remote server ;). there's no X, no gui, no icons to click. at best, there's some kind of web interface.
I've never even administered a local server. The only servers I work with are the 2 legged ones at restaurants and they've usually been setup to administer themselves, just hold out the prospect of a good tip and sit back and enjoy. I thought I would try to help anyway since nobody else seemed to be jumping at the opportunity and it seemed like a potential learning experience for me too :)
i think i'll pass. the whole idea was to donate some spare cpu cycles to the advancement of science with minimum fuss. it's bad enough that abc@home is using a closed source binary, which is doing god-knows-what. i'm not gonna risk putting it on a production server.
Well, never gamble what you can't afford to lose but you seemed to think it was a good idea at one time. If it's just the problems and frustration you've encountered so far then give it another think and keep asking questions... plenty of noobs with less Linux experience than you have made it work with not too much trouble and you can too. Plenty of help available here, just ask.
i thought it might be a good idea because (a) there are servers with plenty of spare cpu cycles and they are all running 64bit linux; (b) abc@home greatly benefits from 64bit platforms, and (c) on gentoo it just works(tm). ok, i had to email abc people asking how to register an account using command line, but their reply was quick and very helpful.
on debian, what should have been a 5-minute job, became several hours. and i don't think debian is to be blamed, rather boinc and abc developers.
Not ABC devs. Their binary runs fine if you have a working BOINC installation and it's properly configured. You didn't have that.
BOINC devs? It would be nice if they had the time to build a package that slides nicely into all the different Linux distros but they have to make do with what they have. A lot of the expertise they have is volunteered, as you may already be aware.
for anyone interested, the problem was resolved after upgrading boinc-client to 5.4.11-4 from debian backports <http://www.backports.org/dokuwiki/doku.php?id=instructions>.
Joy at last :) Just one small warning here... 5.4.x may work well enough at ABC but I think a few other projects might require 5.8.x... just in case you want to do other projects in the future.
for the abc@home project manager(s):
if you are interested in attracting cpu time from linux servers, you have to make sure that:
1) your project works out of the box on boinc cluents included in major server distros (the ones with long term support). this means rhel and it's clones (centos, white box, etc), novell/suse, debian stable and ubuntu LTS.
I think they're interested in attracting cpu time from people who have a working BOINC installation. For those who don't there are the BOINC forums, forums for their distro, etc.
2) minimum system requirements, including version of boinc client supported, are clearly posted on the front page.
Agreed, very important.
3) (link to) instructions on how to jump start the project using command line client clearly posted on the the front page.
Front page clutter issue there.
4) open source abc app for security audit.
You've obviously never done much research or administered a BOINC project ;)
|
|
|
|
|
you found the answer and in fact found the same on a German website proposing to try a newer distro. http://www.boinc-team.de/portal/showtopic.php?threadid=1277
sorry, i don't speak german. and trying another distro is not an option.
|
|
|
|
|
for the abc@home project manager(s):
if you are interested in attracting cpu time from linux servers, you have to make sure that:
1) your project works out of the box on boinc cluents included in major server distros (the ones with long term support). this means rhel and it's clones (centos, white box, etc), novell/suse, debian stable and ubuntu LTS.
I think they're interested in attracting cpu time from people who have a working BOINC installation. For those who don't there are the BOINC forums, forums for their distro, etc.
ok, let me explain you. suppose you're a sysadmin. you have a number of servers that have plenty of spare cpu cycles (to put things into perspective, each of them can do several thousand "credits" worth of work per day, just look at the top hosts list). all of these servers are sitting either in the server room with limited access or at datacenter(s) half way around the globe. most of them are running 64 bit. all of them are there to perform other important jobs, not to run boinc.
you know nothing about boinc or abc. you've read in mainstream computer press that there's a way to donate your spare cpu cycles to science and there's one this project in particular that is trying to solve an important mathematical problem and it greatly benefits from 64 bit processing. you check repos for your distro and bingo -- that boinc thingy is there. you go to your boss and surprise -- he's open minded enough to ok it as long as it doesn't compromise security or disturb operation of the servers.
now, you issue aptitude install boinc-client and it went fine. you go to http://abcathome.com/ and read 'When prompted, enter http://abcathome.com/'. nothing was prompted. you google for 'debian boinc how-to' and get http://wiki.debian.org/BOINC/Troubleshooting, follow it and get where i've started this thread.
and yeah, keep in mind that you can't post to this forum without creating an account first and there're no way of creating an account online.
that's not how it should be.
3) (link to) instructions on how to jump start the project using command line client clearly posted on the the front page.
Front page clutter issue there.
see above. one link won't create any clutter and without this link you're in catch 22.
4) open source abc app for security audit.
You've obviously never done much research or administered a BOINC project ;)
research is what i do for living. and the only reason for not opening source i can imagine is that they've incorporated some proprietary code to save time. |
|
|
|
|
for the abc@home project manager(s):
if you are interested in attracting cpu time from linux servers, you have to make sure that:
1) your project works out of the box on boinc cluents included in major server distros (the ones with long term support). this means rhel and it's clones (centos, white box, etc), novell/suse, debian stable and ubuntu LTS.
I think they're interested in attracting cpu time from people who have a working BOINC installation. For those who don't there are the BOINC forums, forums for their distro, etc.
ok, let me explain you. suppose you're a sysadmin. you have a number of servers that have plenty of spare cpu cycles (to put things into perspective, each of them can do several thousand "credits" worth of work per day, just look at the top hosts list). all of these servers are sitting either in the server room with limited access or at datacenter(s) half way around the globe. most of them are running 64 bit. all of them are there to perform other important jobs, not to run boinc.
you know nothing about boinc or abc. you've read in mainstream computer press that there's a way to donate your spare cpu cycles to science and there's one this project in particular that is trying to solve an important mathematical problem and it greatly benefits from 64 bit processing. you check repos for your distro and bingo -- that boinc thingy is there. you go to your boss and surprise -- he's open minded enough to ok it as long as it doesn't compromise security or disturb operation of the servers.
now, you issue aptitude install boinc-client and it went fine. you go to http://abcathome.com/ and read 'When prompted, enter http://abcathome.com/'. nothing was prompted. you google for 'debian boinc how-to' and get http://wiki.debian.org/BOINC/Troubleshooting, follow it and get where i've started this thread.
and yeah, keep in mind that you can't post to this forum without creating an account first and there're no way of creating an account online.
that's not how it should be.
Well I pretty much figured it was that way for you and I understand and sympathize because I have been in very similar situations (just not related to servers) so many times in the past and know thousands with similar experiences.
I agree that's not the way it should be. In a better world projects like ABC would have all the funds they need to hire all the people they need to give us crunchers everything we need to make life easy. And the volunteers building packages for ALL the distros and even Microsoft would have solid drop in installers that just work. In an even better world, taxes would be spent on buying supercomputers for researchers instead of making war and we wouldn't need to crunch at all. Alas, we have Taliban, Bush, AlQaeda and at least a dozen other arseholes calling the shots, so we do what we can.
All I can say is... your efforts to get it working and your posts are appreciated, tell your boss s/he is a good person.
3) (link to) instructions on how to jump start the project using command line client clearly posted on the the front page.
Front page clutter issue there.
see above. one link won't create any clutter and without this link you're in catch 22.
There's 22 or more catch 22s. 22 links? But I agree in principle, better docs and help is sorely needed. Again it's a matter of manpower.
4) open source abc app for security audit.
You've obviously never done much research or administered a BOINC project ;)
research is what i do for living. and the only reason for not opening source i can imagine is that they've incorporated some proprietary code to save time.
Seems like proprietary code is basically the issue. So why don't they just get over their selfish streak and share? Well, I see at least one very good reason related to integrity of the results and I imagine that's just as important an issue to them as security is to you.
|
|
|