|
Questions and Answers :
Unix/Linux :
*** Running 32bit CPDN from 64bit Linux - Discussion ***
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 18 · Next
Author | Message |
---|---|
![]() Send message Joined: 31 Dec 07 Posts: 1152 Credit: 22,140,554 RAC: 418 |
To expand on what Dave said. So long as the UK Met Office can take advantage of the fact that the 32 bit models run on 64 bit Windows right out of the box, and the fact that there are so many more Windows computers out there for them to run the models on, it is unlikely that they will ever go to the time and expense of producing a 64 bit Linux (or Windows or Mac) version. |
![]() Send message Joined: 7 Aug 04 Posts: 2172 Credit: 64,684,134 RAC: 794 |
It's not terribly hard to configure Linux with 32 bit support (at least on Ubuntu and related distros), therefore able to run CPDN tasks. But apparently it's difficult enough to cause big problems running models on Linux. Well, it's just that there are so many people who just attach and forget it. They never look at their webpage results output, or the message log in boinc. They just assume that if they attach to a project, it's going to work. Obviously lots and lots of people are like this, and on many projects, it works that way. Here...not so much. No secret nearly all Linux applications these days are 64 bit, which leaves me curious why CPDN Linux models cling to the older 32 bit design. Sure seems you are right, it's not at all using the power of Linux computers to their potential on behalf of CPDN. I don't believe it is a huge deal to compile for 64bit vs. 32bit. There is not a whole rewrite of the code needed, but perhaps some changes might need to be made. I think it is more to do with alpha/beta testing and validation time and effort. The results output for the same task may not be bit for bit identical, but then it is also not for Windows vs. Linux with the different platforms and optimizations, or even AMD and Intel processors of various generations. And we've been running that way for many years with any differences in output about equivalent to initial condition perturbations in ensemble output. |
Send message Joined: 4 Jul 15 Posts: 63 Credit: 3,223,760 RAC: 0 |
Honestly I can't understand the "start and forget" attitude some users evidently have. I tend to watch my systems like a hawk, and wouldn't dream of doing otherwise. We're talking about computing after all, things notoriously go wrong all the time. Especially with systems like Linux that aren't "off the shelf" like Windows, rather surprising that anyone geek-minded enough to bother with it would neglect how their machines are handling CPDN tasks. A while back, out of curiosity I counted the number of Linux installations there were among the "top 100" CPDN computers. There were a lot more than I expected, 14/100 if memory serves correctly. It struck me as quite a significant chunk of the computer resources serving CPDN, thus a reason to try to keep Linux systems crunching for the project's benefit. If converting to 64 bit applications would reduce wasted effort and speed up results, it seems likely to be worth it given Linux is hardly a rarity out here. |
![]() Send message Joined: 6 Aug 04 Posts: 264 Credit: 965,476 RAC: 0 |
Of my six BOINC projects only ATLAS@home and CMS-dev require a 64-bit Linux. All others use 32-bit Linux apps. My two Linux boxes run 32-bit OS, although one has a 64-bit CPU. My 64-bit Windows 10 CPU runs three 64-bit Linux Virtual Machines, one 64-bit Solaris Virtual machine and one 32-bit Virtual Machine, vLHC@home. CERN has said it will update vLHC@home but so far it has not happened. The CERN 2015 Challenge, not a BOINC project. uses 64-bit Linux Virtual Machine. I am running a CPDN task on this SuSE Linux 32-bit host, plus vlHC@home. SETI@home runs 32-bit on my other Linux box, together with Einstein@home. Both SETI@home and Einstein@home run on the 64-bit Windows 10 PC. Tullio |
![]() Send message Joined: 15 May 09 Posts: 4444 Credit: 17,929,339 RAC: 29,974 |
For completeness, I hadn't bothered looking at getting graphics working since upgrading to latest ubuntu as they weren't available on the tasks I was running anyway. Now there are tasks with graphics, on my box, sudo apt-get install libjpeg62:i386 Got them working. |
Send message Joined: 24 Aug 08 Posts: 7 Credit: 37,536,564 RAC: 163 |
What can cause the message Download Failed? There is plenty of free space on the disk. |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
Look in the Stderr list on the page for that model. Of more concern, is that some of your computers are crashing everything, possibly due to a permissions problem. |
![]() Send message Joined: 15 May 09 Posts: 4444 Credit: 17,929,339 RAC: 29,974 |
The download failed ones are because a file is in the wrong place on the server if I remember correctly. |
Send message Joined: 4 Jul 15 Posts: 63 Credit: 3,223,760 RAC: 0 |
I had two tasks fail with the same download error that Wes reported. If I understand what's going on, the server wasn't able to find >=1 file the client requested, though not necessarily the same files on repeated tries. I'm guessing this error is affecting more than a few clients. The fact that for the same work unit successive clients receive download errors for different sets of files is a curiosity, though the important thing appears to be that quite a few files intended to be downloaded aren't in the proper location. Is it fair to call this a problem of server misconfiguration? |
Send message Joined: 5 Sep 04 Posts: 7629 Credit: 24,240,330 RAC: 0 |
There may be a couple of "errors" that apply in this case: 1) Some files for one of the several separate batches of work, were needed to be in a location that didn't have the correct permissions; once they were in place, they couldn't be read. 2) One of the several separate batches of work had an error in the greenhouse gases provided for them. All of this was fixed soon after being reported to the project. But the server software is such that once the model and it's auxiliary files has been grabbed by someone, those can't be pulled from the queue. So only those that HADN'T been issued were pulled. Then the whole batch was re-issued. For the rest, it's just a matter of waiting for a computer to grab one of the faulty sets, and thus reduce it's "Number of re-tries", and hope that it's not one of your computers. |
Send message Joined: 4 Jul 15 Posts: 63 Credit: 3,223,760 RAC: 0 |
Thanks for the information, interesting, I hadn't considered permissions, then again I'd think "read" access would be available for a file in the right directory. More relevant is the history of the two "fail to download" tasks. In both instances the tasks had been previously crunched until "error while computing" (exit code 193), followed by download errors on being sent to two computers on Dec 5 2015, my computer being the last. (Re: workunits 9887820 and 10008664.) It wasn't clear when the problem was fixed, maybe after Dec 5. In that case it looks like I was "unlucky" enough to get the bum tasks. At least it seems my systems were not the source of the issue, well, this time anyway. Thanks, JRA |
![]() Send message Joined: 15 May 09 Posts: 4444 Credit: 17,929,339 RAC: 29,974 |
On my laptop, sudo apt-get install wine followed by downloading the boinc .exe file and clicking on it is straightforward. Just need to make sure that the linux version of boinc isn't running or the windows version doesn't work. Same happens the other way around on my boxes at least, the linux version doesn't connect if the other one is running under WINE. I now have 4 of the latest batch of WAH2_eu tasks running all over an hour in with no problems so far. Will post later to confirm that uploads etc work normally. I read that there were minor problems with the processor type being reported incorrectly but that seems to have been resolved with the latest wine and boinc. |
![]() Send message Joined: 15 May 09 Posts: 4444 Credit: 17,929,339 RAC: 29,974 |
First zips all gone without issue. took a bit of searching to find the projects folder for the wine installation but got there in the end. Looks like so far no nasty surprises. |
Send message Joined: 8 Aug 04 Posts: 69 Credit: 1,561,341 RAC: 0 |
sudo apt-get install libxmu:i386 gets the graphics to work, at least it does on my beta tasks, none of the other work I have has graphics available. E:unable to locate package seems this lib is obsolete now. Can You help me with another package name? Tnx :) ChrisD |
![]() Send message Joined: 15 May 09 Posts: 4444 Credit: 17,929,339 RAC: 29,974 |
which distribution are you using? it should be possible if unavailable for yours to find another .deb or .rpm package which you can download and install. It may be that none of the new tasks will have graphics so this may be academic. |
Send message Joined: 3 Mar 13 Posts: 2 Credit: 13,423,511 RAC: 0 |
I used to do this when I ran Lubuntu, and it always worked. I found the information somewhere on the internet, but I don't remember where. 1. sudo apt-get install lib32z1 lib32ncurses5 lib32bz2-1.0 libstdc++6 libstdc++5 freeglut3 2. You also need from a 32 bit system libgcc_s.so.1 and libstdc++.so.6 Copy and paste those two files into /lib32 . I believe I got those files from a 32 bit Lubuntu distribution which I now keep on a USB flash drive. |
![]() Send message Joined: 15 May 09 Posts: 4444 Credit: 17,929,339 RAC: 29,974 |
1. sudo apt-get install lib32z1 lib32ncurses5 lib32bz2-1.0 libstdc++6 libstdc++5 freeglut3 That on it's own doesn't satisfy the dependencies for the graphics to work. E:unable to locate package A bit surprised as it worked for me on the latest xubuntu distribution. If it doesn't get installed automagically by one of the others I think you may also need sudo apt-get install libjpeg:i386 |
Send message Joined: 1 Sep 04 Posts: 161 Credit: 81,459,123 RAC: 856 |
I found on some thread (not related to BOINC or CPDN) on the Internet in a reply to a reply that installing libstdc++6:i386 might solve missing libstdc++.so.6 issues. It seems to have done just that for me. There has to be a better way of determining if a computer has the necessary libraries installed than downloading 10 CPDN tasks and having them all abort. For the not so expert LINUX user, clear, unambiguous instructions on how to resolve missing libraries would be nice too. |
![]() Send message Joined: 15 May 09 Posts: 4444 Credit: 17,929,339 RAC: 29,974 |
There is, go to the projects/climateprediction.net directory and in a terminal type ldd executable name and you will get an output something like ldd hadam3p_eu_7.23_i686-pc-linux-gnu linux-gate.so.1 => (0xf7715000) libpthread.so.0 => /lib/i386-linux-gnu/libpthread.so.0 (0xf76c5000) libdl.so.2 => /lib/i386-linux-gnu/libdl.so.2 (0xf76c0000) libstdc++.so.6 => /usr/lib/i386-linux-gnu/libstdc++.so.6 (0xf7548000) libm.so.6 => /lib/i386-linux-gnu/libm.so.6 (0xf74fb000) libgcc_s.so.1 => /lib/i386-linux-gnu/libgcc_s.so.1 (0xf74dd000) libc.so.6 => /lib/i386-linux-gnu/libc.so.6 (0xf7322000) /lib/ld-linux.so.2 (0x565cc000) As all the dependencies for this executable are met it doesn't show any unmet ones but it will show you what needs to be installed. Sometimes you may need to do some research to find which package contains the missing libs. Yes it would be nice if there was a command to find and install them but if there is a command/script to do that I haven't found it yet. |
Send message Joined: 1 Sep 04 Posts: 161 Credit: 81,459,123 RAC: 856 |
Thanks Dave - That addresses 90% of my concerns. I installed (upgraded) to 64-bit UBUNTU a few months ago but haven't been able to use it on CPDN because of the because of the absence of work. I had about 10 tasks download and abort before I realized what was happening. If you have a "new" installation or maybe a new .gnu files from CPDN, you would have to - 1) Download at least one task and let it abort to get the new .gnu file(s) 2) Run the ldd command to find missing libraries 3) Install some stuff 4) Repeat steps 2 and 3 until nothing missing That should work. I wonder what percentage of the CPDN LINUX users know this? |
©2024 climateprediction.net